Q: How to Narrow My Options and Move Forward?

Q: I have been trying to narrow down the number of options I am considering for my startup and pick a clear direction. But the more I try to refine my loose plans into a clear strategy, the harder and more draining the effort feels. Why does that happen? What suggestions do you have for converging a set of options into a workable plan and start moving forward?

A: I think some people really like to hold on to possibilities, and others like to see things settled and have a definite plan. My sense is that you enjoy having a range of options and value optionality more than having a definite plan you can execute. Please note that this is different from crafting and executing a plan with branches and sequels, because in the latter case you are moving forward. The net effect of not making a choice is that you will let events foreclose options or may fail to do much of anything–leaving “men on base” instead of actually focusing and completing a few projects.

Define a sampling strategy where you execute scale model or prototype versions and see if they are viable. Or determine which options will expire first. If you don’t want to lose them, then you should get started. If you delay, you will have allowed them to lapse. Of course, new possibilities may have occurred to you by then, giving you a rich range of options to leave unexplored except in your mind.

It’s certainly possible to decide to quickly, engaging in “extinction by instinct.” But you current approach seems to leave you trapped in “analysis paralysis.”

Q: Some people really do seem wired to settle and move forward, while others stay open longer because the possibilities themselves feel alive and worth protecting. I have a good job but I have this dream of doing a startup, but I don’t want to screw it up.

This leads to a paralysis isn’t just indecision, it’s the sheer mental cost of keeping the whole map in view while still trying to narrow anything down. The “men on base” feeling is real, but sometimes those runners are the only thing keeping the bigger picture from collapsing into something too small too fast.

I have been trying to write everything down just so my brain doesn’t have to keep track of every branch, but I find I keep revising and extending the possibilities and I am  not converging.

A: How much difference is there between the options you are considering? Unless this is a substantial life decision, you are better served by picking a direction and taking a small step forward. Action will tell you more than any amount of mental simulation.

You don’t have to quit your job to start interviewing prospects or working on a product nights and weekends. In fact I would encourage not to quit your job until you are sure a startup is something you want to do and you have evidence that people will pay for what you have in mind.

One way to get some clarity, if writing it down is not helping, would be to sit down with a trusted friend and walk around the situation with them. The one rule is that they can only ask clarifying questions unless you specifically ask them for advice. Record the conversation and listen to it later or look at the transcript to hear yourself. This is a scaled down version of what the Quakers call a “Clearness Committee” I wrote about this in “Asking Questions from a Caring Perspective” See for a more detailed description. You are also welcome to attend a Bootstrappers Breakfast and talk about one of your ideas, sometime the act of voicing your idea can help you organize your thoughts.

I never want to tell someone they should not pursue their dreams or they cannot accomplish something, but it is a source of concern that you want to develop a perfect plan and then execute it. It’s a very rare occasion you can do this as an entrepreneur. You make your plan when you have the lease amount of information and then will have to update it based on what you learn and events that occur.

Most options in any decision don’t lead to significantly different outcomes, or at least differences that are foreseeable at the time of taking action. The loss of time is almost always the more critical resource than trying to optimize a choice unless it’s a significant decision (e.g., one that comes up less than once a year).

I enjoy playing turn-based computer games that are complex. Unfortunately this bears little resemblance to what is needed to get a startup off the ground. Most opportunities are perishable and you need to act before you have all of the information you would like. You need to continually update your plans in response to changes in the market. It’s not like building scenery for your model railroad or collecting stamps where you can work at your own pace and when you feel like it. I think you should try to talk to some other founders about what their first two years were like and how much their plans changed.

Related Blog Posts

• Uncertain Times
• Uncertain Times Are The End of An Illusion
• Dr. Atul Gawande on Managing Complexity and Uncertainty
• Making Business Decisions in Uncertain Times
• Planning in a Bootstrapped Startup: a Model from Will Kamishlian
• A Holistic Approach to Launching a Bootstrapped Startup
• Chalk Talk: Write Down Key Hypotheses

Quotes for Entrepreneurs Curated in July 2026

I curate these quotes for entrepreneurs from a variety of sources and tweet them on @skmurphy about once a day where you can get them hot off the mojo wire. At the end of each month I curate them in a blog post that adds commentary and may contain a longer passage from the same source for context.

My theme for this month’s “Quotes for Entrepreneurs” is artificial intelligence.

+ + +

“Much to the surprise of the builders of the first digital computers, programs written for them usually did not work.”
Rodney Brooks in “Programming in Common LISP” (1985)

+ + +

“AI is a tool.
But you need to provide the goal.”
Balaji Srinivasan (blog https://balajis.com/) in a Feb-18-2026 tweet

+ + +

“What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.”
Joseph Weizenbaum in “Computer Power and Human Reason“

Weizenbaum wrote ELIZA (original 1966 CACM article), a program that mimicked a Rogerian therapist but had no true intelligence, just some simple open ended questions triggered by keywords in user text input.For example mentioning mother or father would trigger “Tell me about your parents.” Here is a longer excerpt to put the quote in context (bold added)

“[P]eople were conversing with the computer as if it were a person who could be appropriately and usefully addressed in intimate terms. I knew of course that people form all sorts of emotional bonds to machines, for example, to musical instruments, motorcycles, and cars. And I knew from long experience that the strong emotional ties many programmers have to their computers are often formed after only short exposures to their machines. What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people. This insight led me to attach new importance to questions of the relationship between the individual and the computer, and hence to resolve to think about them.”

Joseph Weizenbaum in “Computer Power and Human Reason“

In the book, Weizenbaum outlines three surprises he experienced from mistaken beliefs the program generated. I think these are true today of modern LLMs:

1. A number of practicing psychiatrists seriously believed the DOCTOR computer program could grow into a nearly completely automatic form of psychotherapy.
2. I was startled to see how quickly and how very deeply people conversing with DOCTOR became emotionally involved with the computer and how unequivocally they anthropomorphized it.
3. Another widespread, and to me surprising, reaction to the ELIZA program was the spread of a belief that it demonstrated a general solution to the problem of computer understanding of natural language.

+ + +

“If you want to become AI native you have to start small and compound wins over time. Start by automating workflows that have measurable ROI then expand from there.”

Collin McClellan, CEO at Collide.io, in a May-27-2026 tweet

+ + +

“As we come to rely on computers to mediate our understanding of the world, it is our own intelligence that flattens into artificial intelligence.”
Nicholas Carr in “The Shallows: What the Internet Is Doing to Our Brains” (2008)

+ + +

“Stated simply, the techniques of artificial intelligence are to the mind what bureaucracy is to human social interaction. […] The benefits of bureaucracy follow from the reduction of judgment to the systematic application of explicitly articulated rules. Bureaucracy achieves a predictability and manageability that is missing in earlier forms of organization.

There are striking similarities here with the arguments given for the benefits of expert systems and equally striking analogies with the shortcomings as pointed out, for example, by March and Simon in Organizations:

The reduction in personalized relationships, the increased internalization of rules, and the decreased search for alternatives combine to make the behavior of members of the organization highly predictable; i.e., the result in an increase in the rigidity of behavior of participants increases the amount of difficulty with clients of the organization and complicates the achievement of client satisfaction.

Terry Winograd in Thinking Machines: Can There Be? Are We? collected in The Boundaries of Humanity: Humans, Animals, Machines

+ + +

“Understand where human judgement adds value: tasks that require experience, commercial judgement, or relationship building should not be automated. Measure operational outcomes rather than technical success.”
Jan Rautenbach (@JanRautenbach7) in “What Can We Automate?“

This reminds me of Rory Sutherland’s “Doorman fallacy”

When you pursue efficiency, you generally start by looking at numerical or mechanical factors. In the process, you tend to disregard psychological factors, where the greater gains are often found. As a result, you focus too heavily on cost reduction and too little on value creation. One of the greatest forms of efficiency, by the way, is employing a human being who’s really nice.

Imagine a five-star hotel with a doorman who welcomes incoming guests. A management consulting firm comes in and says, “Your doorman currently costs you $60,000 a year. We’ve defined his or her function as opening the door. We’ll replace the doorman with an automatic door mechanism and an infrared sensor, saving you $30,000 to $40,000 a year.”

They walk away, take credit for the cost saving, and two years later the hotel is in trouble. The rack rate has fallen off a cliff because the doorman was actually doing many things—most of them human and tacit. Security, for example—keeping vagrants from sleeping in the doorway—hailing taxis, dealing with luggage, recognizing regular guests, and providing status to the hotel. There are many value-creation components to that doorman that aren’t captured in the narrow “open the door” definition. It’s a clear example of where the costs are highly visible, but the benefits are hard to see.

As Roger L. Martin says, “Any idiot can cut costs; what takes real skill is cutting costs in a way that doesn’t destroy value.” The human component—the face-to-face component—does really heavy lifting in any business or experience.”

Rory Sutherland in “Doorman fallacy“

+ + +

“Every major civilization changing development in the last half century, like GPS, the Internet and AI has passed through three stages:

1. It begins as an evil invention of the US military industrial complex;
2. It becomes so indispensable to the world that living without it is unthinkable.
3. It finally grows too important to be left in the hands of its inventors and calls grow for its control by the United Nations.

I predict that access to low earth orbit and the exploitation of space resources will follow the same path as GPS, Internet and AI. They are a “waste of money that could be spent on welfare” now but one day they will be a ‘human right’ whose provision must be guaranteed by the United Nations.”

Richard Fernandez (@WretchardTheCat) in July-6-2026 tweet that linked to “Global push for AI governance amid warnings of ‘catastrophic harm’“

+ + +

“The official story is incomplete, and here’s what I think it leaves out. My selection is also incomplete. I have biases, incentives, and blind spots. I’m giving you my read, not the truth. You’re going to have to figure this out yourself.”
Mark Atwood in “Disillusionment is Also a Product“

It would be nice of AI CEOs, AI commentators and pundits, AI VCs, and other AI experts all spoke like this.

+ + +

“AI has been compared to various historical precedents: electricity, industrial revolution, etc., I think the strongest analogy is that of AI as a new computing paradigm (Software 2.0) because both are fundamentally about the automation of digital information processing.

If you were to forecast the impact of computing on the job market in ~1980s, the most predictive feature of a task/job you’d look at is to what extent the algorithm of it is fixed, i.e. are you just mechanically transforming information according to rote, easy to specify rules (e.g. typing, bookkeeping, human calculators, etc.)? Back then, this was the class of programs that the computing capability of that era allowed us to write (by hand, manually).

With AI now, we are able to write new programs that we could never hope to write by hand before. We do it by specifying objectives (e.g. classification accuracy, reward functions), and we search the program space via gradient descent to find neural networks that work well against that objective. This is my Software 2.0 blog post from a while ago. In this new programming paradigm then, the new most predictive feature to look at is verifiability. If a task/job is verifiable, then it is optimizable directly or via reinforcement learning, and a neural net can be trained to work extremely well. It’s about to what extent an AI can “practice” something. The environment has to be resettable (you can start a new attempt), efficient (a lot attempts can be made), and rewardable (there is some automated process to reward any specific attempt that was made).

The more a task/job is verifiable, the more amenable it is to automation in the new programming paradigm. If it is not verifiable, it has to fall out from neural net magic of generalization fingers crossed, or via weaker means like imitation. This is what’s driving the “jagged” frontier of progress in LLMs. Tasks that are verifiable progress rapidly, including possibly beyond the ability of top experts (e.g. math, code, amount of time spent watching videos, anything that looks like puzzles with correct answers), while many others lag by comparison (creative, strategic, tasks that combine real-world knowledge, state, context and common sense).

Software 1.0 easily automates what you can specify.

Software 2.0 easily automates what you can verify.”

Andrej Karpathy in Nov 16, 2025 tweet

+ + +

“Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.”
Vernor Vinge  Opening sentence for “The Coming Technological Singularity: How to Survive in the Post-Human Era (1993)

+ + +

“Observe the wonders as they occur around you.
Don’t claim them. Feel the artistry
moving through, and be silent.”

Rumi in “Body Intelligence”

Collected The Essential Rumi (1995) translated by Coleman Barks, this verse occurs on page 152. I thought this captured raw intelligence embedded in our bodies and the wonders, natural and man-made, that occur around us every day.

An interesting book on “body intelligence” is “The Future of the Body” (1992) by Michael Murphy (no relation). One concept he explores in the book is integrative embodiment where capabilities are woven by regular practice into a transformed style of being rather than rare or peak experiences.

One example developed by Joseph Heinrich in “The WEIRDest People in the World” is literacy (drawing on “Inside the Letterbox: How Literacy Transforms the Human Brain” by Stanislas Dehaene) is that the daily practice of reading has profound effects on our brain structure and how we process our visual field. This plasticity of our brains strikes me as a two-edge sword: it can make us weak and lazy (see for example Superstimulus: Refining Online Interactions into Digital Heroin) or extend our capabilities, as suggested by Vernor Vinge:

“Computer/human interfaces may become so intimate that users may reasonably be considered superhumanly intelligent.”
Vernor Vinge  in”The Coming Technological Singularity: How to Survive in the Post-Human Era (1993)

This is one of several alternate paths to a Technological Singularity that Vinge postulates. It’s interesting that dyslexics inherit a slightly different brain structure that enables better three-dimensional visualization and much less right-left preference (closer to equal skill with both hands or both foot) that would have had significant Stone Age survival value but places them at a small disadvantage in a literate culture without specialized training in phonics to enable phonemic awareness.

+ + +

“A computer can never be held accountable, therefore a computer must never make a management decision.”
Slide from a 1979 IBM Training Session  entitled “The Computer Mandate”

I fear that the value of this injunction will be rediscovered several times as some rush to put increasingly opaque algorithms in charge of increasingly important aspects of our lives. I originally curated this in March 2024 but it seems a good fit with a lot of AI initiatives.

+ + +

“I’m calling it now, the adoption of AI agents into software development will be one of the most costly mistakes in the field’s history. Agents cannot program, and it’s taking longer and longer to realize that they can’t. They are a highly sophisticated statistical model designed to mimic the distribution of programming. The output is broken, but in a way that’s getting harder and harder to detect. Which is exactly what you’d expect from an increasingly accurate statistical model.”

George Hotz (@RealGeorgeHotz) in “The Eternal Sloptember“

It’s an interesting perspective. Title is a riff on “Eternal September”

+ + +

“It’s funny, people that have the strong desire to plan everything in advance and prevent any problems can provide a lot of value, but are typically like antimatter in an early stage startup where getting started, improvising, listening, learning and adjusting are what is required.”
Sean Murphy in a March-2021 email to another Bootstrappers Breakfast moderator

I was probably influenced by a model Paul Saffo had proposed for this in “Ready Fire Steer.” He also cautions “never mistake a clear view for a short distance.” I think we are at the “Fall down 7 times, stand up 8” stage of exploring the new possibilities AI has unlocked.

+ + +

The assumptions behind a superhuman intelligence arising soon are:

1. 1. Artificial intelligence is already getting smarter than us, at an exponential rate.
   2. We’ll make AIs into a general purpose intelligence, like our own.
   3. We can make human intelligence in silicon.
   4. Intelligence can be expanded without limit.
   5. Once we have exploding superintelligence it can solve most of our problems.

In contradistinction to this orthodoxy, I find the following five heresies to have more evidence to support them.

1. Intelligence is not a single dimension, so “smarter than humans” is a meaningless concept.
2. Humans do not have general purpose minds, and neither will AIs.
3. Emulation of human thinking in other media will be constrained by cost.
4. Dimensions of intelligence are not infinite.
5. Intelligences are only one factor in progress.

Kevin Kelly in “The Myth of Superhuman AI“

+ + +

“Let’s make a checklist to help in the discourse on public decision-making. Should one not ask of any public project or loan whether it:

1. promotes justice;
2. restores reciprocity;
3. confers divisible or indivisible benefits;
4. favors people over machines;
5. whether its strategy maximizes gain or minimizes disaster;
6. whether conservation is favored over waste; and
7. whether the reversible is favored over the irreversible? The last item is obviously important.

Considering that most projects do not work out as planned, it would be helpful if they proceeded in a way that allowed revision and learning, that is, in small reversible steps.

Ursula Franklin in The Real World of Technology page 126 in (1990)

Seems useful for large AI projects as well. Nothing is truly reversible, so it may be better framed as starting at a small scale and having mitigation and shutdown plans prepared in case the project impacts deviate substantially from the anticipated trajectory. Please note that this is not a stealth argument on my part against data centers; I am much more concerned about regulatory capture freezing the evolution the same way that heavy regulation has slowed down health care improvements and defended slow-moving incumbents from competition that would have yielded substantial improvements in patient outcomes.

+ + +

“The man of the hour has an air of great power.”
Curtis Mayfield  in “Superfly”

This seems to me to be the public personas of Sam Altman or Dario Amodei today, and Bill Gates or Mark Zuckerberg when they were ascendant. Elon Musk has a slightly different set of concerns:

“With artificial intelligence, we are summoning the demon. There are all those stories where a guy is sure he can control a demon with a pentagram and holy water, but it doesn’t work out.”

Elon Musk at MIT Aeronautics and Astronautics Department’s Centennial Symposium (October 24, 2014)

Musk’s remarks reminded me of a scene in Neuromancer where the “Turing Police” attempt to detain Case and Michle makes this short speech.

“You are worse than a fool, you have no care for your species. For thousands of years men dreamed of pacts with demons. Only now are such things possible. And what would you be paid with? What would your price be, for aiding this thing to free itself and grow?”
William Gibson in Neuromancer (1984) Chapter 13 confrontation with Turing Police

+ + +

“Arguing against data centers is like arguing against the existence of the Internet itself. We’re not putting that genie back into that bottle.” Jim Geraghty in Why the Data Center Construction Boom Is Hitting a Wall

A detailed non-technical analysis of what’s required to build a data center and why they are important. His concluding paragraphs:

“Add it all up, and building a new data center requires a location that is preferably not in the middle of nowhere, with access to a massive amount of electricity and water, with space for major power and coolant components, in a building either built to specifications or likely heavily overhauled from its intended use, with major physical and cybersecurity, built by a limited pool of skilled labor, and with a reliable supply for all the transformers, switchgear, and batteries inside.

Considering all this, it’s kind of amazing that any data center ever gets built. Notice I have not written, “Therefore, we should not build data centers.” I am simply laying out the major logistical challenges and pointing out why we should not be surprised to see the rate of data center construction slow down. […] A better and more productive discourse would focus on cold, hard facts like the ones listed above.

Jim Geraghty in Why the Data Center Construction Boom Is Hitting a Wall

 + + +

“The AI risk for SaaS isn’t that customers will build their own but that the barrier to entry for competitors is lower.

The Chorleywood process created mega bakeries that displaced regular bakeries because they changed the economics. AI is doing the same and fundamentally changing the economics of production. What used to take years and huge teams to build can be built by much smaller teams much faster.

SaaS isn’t going to sublimate straight into consumer built tools but the boiling point for competition has gotten a lot lower.”

HN user NotARobot123 in Comment https://news.ycombinator.com/item?id=48916828

+ + +

“Man-computer symbiosis is an expected development in cooperative interaction between men and electronic computers. It will involve very close coupling between the human and the electronic members of the partnership. The main aims are 1) to let computers facilitate formulative thinking as they now facilitate the solution of formulated problems, and 2) to enable men and computers to cooperate in making decisions and controlling complex situations without inflexible dependence on predetermined programs. In the anticipated symbiotic partnership, men will set the goals, formulate the hypotheses, determine the criteria, and perform the evaluations. Computing machines will do the routinizable work that must be done to prepare the way for insights and decisions in technical and scientific thinking. Preliminary analyses indicate that the symbiotic partnership will perform intellectual operations much more effectively than man alone can perform them.”

J. C. R. Licklider in the abstract for Man-Computer Symbiosis (1960)

This is a clear view from 1960 of where we are today. Originally published in IRE Transactions on Human Factors in Electronics, volume HFE-1, pages 4-11, March 1960

+ + +

Licklider  also includes a very interesting “time and motion” analysis of his own technical thinking in the same paper. Here is a summary:

“About 85 per cent of my “thinking” time was spent getting into a position to think or learn something I needed to know. Much more time went into finding or obtaining information than into digesting it. […] My “thinking” time was devoted mainly to activities that were essentially clerical or mechanical and my choices of what to attempt were determined to a great extent by considerations of clerical feasibility, not intellectual capability. […] The operations that fill most of the time devoted to technical thinking are operations that can be performed more effectively by machines than by men. If those problems can be solved in such a way as to create a symbiotic relation between a man and a fast information-retrieval and data-processing machine, the cooperative interaction would greatly improve the thinking process.”
J. C. R. Licklider in section 3.1 (condensed) of  Man-Computer Symbiosis (1960)

Here is full section:

“3.1 A Preliminary and Informal Time-and-Motion Analysis of Technical Thinking

Despite the fact that there is a voluminous literature on thinking and problem solving, including intensive case-history studies of the process of invention, I could find nothing comparable to a time-and-motion-study analysis of the mental work of a person engaged in a scientific or technical enterprise. In the spring and summer of 1957, therefore, I tried to keep track of what one moderately technical person actually did during the hours he regarded as devoted to work. Although I was aware of the inadequacy of the sampling, I served as my own subject.

It soon became apparent that the main thing I did was to keep records, and the project would have become an infinite regress if the keeping of records had been carried through in the detail envisaged in the initial plan. It was not. Nevertheless, I obtained a picture of my activities that gave me pause. Perhaps my spectrum is not typical–I hope it is not, but I fear it is.

About 85 per cent of my “thinking” time was spent getting into a position to think, to make a decision, to learn something I needed to know. Much more time went into finding or obtaining information than into digesting it. Hours went into the plotting of graphs, and other hours into instructing an assistant how to plot. When the graphs were finished, the relations were obvious at once, but the plotting had to be done in order to make them so. At one point, it was necessary to compare six experimental determinations of a function relating speech-intelligibility to speech-to-noise ratio. No two experimenters had used the same definition or measure of speech-to-noise ratio. Several hours of calculating were required to get the data into comparable form. When they were in comparable form, it took only a few seconds to determine what I needed to know.

Throughout the period I examined, in short, my “thinking” time was devoted mainly to activities that were essentially clerical or mechanical: searching, calculating, plotting, transforming, determining the logical or dynamic consequences of a set of assumptions or hypotheses, preparing the way for a decision or an insight. Moreover, my choices of what to attempt and what not to attempt were determined to an embarrassingly great extent by considerations of clerical feasibility, not intellectual capability.

The main suggestion conveyed by the findings just described is that the operations that fill most of the time allegedly devoted to technical thinking are operations that can be performed more effectively by machines than by men. Severe problems are posed by the fact that these operations have to be performed upon diverse variables and in unforeseen and continually changing sequences. If those problems can be solved in such a way as to create a symbiotic relation between a man and a fast information-retrieval and data-processing machine, however, it seems evident that the cooperative interaction would greatly improve the thinking process.”

J. C. R. Licklider in section 3.1 of  Man-Computer Symbiosis (1960)

+ + +

“We can think of no corpus of text more toxic to the training of useful LLMs than the 21st century scientific literature. Imagining a two-by-two matrix with the axes honest vs. dishonest and right vs. wrong, the scientific literature is splashed haphazardly across all four boxes. Papers often belong in multiple quadrants at once, and not always because of the contributions of different authors.”
Dan Recht and Ben Reinhardt in The Scientific Literature is Poisonous to LLMs

It seems to me that the same flaws in the literature that are blocking LLMs from achieving an effective understanding would also be deleterious to human scientists ability to understand what’s real.

+ + +

“Domain knowledge makes you better at using LLMs. […] Human expertise will continue to be useful even as models get stronger. The human is the bottleneck for many tasks, not the model: the difficult part is communicating to the model exactly what kind of solution the human wants. The information is “in the model” already, but it takes a very smart human to pull it out.”
Sean Goedecke in “LLMs Reward Expertise“

+ + +

“Men will handle the very-low-probability situations when they arise. (In current man-machine systems, that is one of the human operator’s most important functions. The sum of the probabilities of very-low-probability alternatives is often much too large to neglect.) Men will fill in the gaps, either in the problem solution or in the computer program, when the computer has no mode or routine that is applicable in a particular circumstance.”

J. C. R. Licklider in section 3.1 of  Man-Computer Symbiosis (1960)

+ + +

“The AI race is usually described as a quest for intelligence, chips, talent, data and geopolitical power. It is all of those things. But underneath, it is something simpler and more general. It is a race to learn what creates value, at what cost, and how to capture it.”
Elijah Eilert in “The AI Race“

+ + +

“The cost function of bad architecture and bad program design cannot be evaluated by running a unit test. The impact hits you three to six months later when you realize the software has become hard to change.”
Dex Horthy (@DexHorthy) in an interview on Pragmatic Engineer at minute 42

+ + +

“Robots and computer-based artificial intelligences are here now. They are, first of all, an Other, an alien intelligence. Everything we know about artificial intelligence so far suggests that what emerges is anything but human-like. Hand-held calculators are already smarter than humans in arithmetic, but it is clear that they are non-human in their thinking. They are as alien in their precision as Spock is. When we succeed in making very smart AIs, these units will be smarter than us in certain ways, but decidedly non-human in their thought patterns. The more AIs we make, the more different and varied they will be in their thinking.”

Kevin Kelly “Nerd Theology” (1999)

+ + +

The Comparative Advantages of Humans and Computers

“Whether humans or computers have the upper hand depends on many factors, including whether the tasks being undertaken are familiar or unique. When tasks are familiar and much data is available, computers will likely beat humans by being data-driven and highly consistent. Although artificial intelligence is advancing rapidly, a general rule of thumb is that when tasks are unique and when data overload is not a problem for humans, humans likely have an advantage. In many situations, the strongest performance comes from humans and computers working together.”
Paul Schoemaker and Phillip Tetlock in “Building a More Intelligent Enterprise”

Originally curated in May 2018; still seems accurate.

+ + +

“All work is the avoidance of harder work.”
James Richardson

It does not seem like folks using AI are working less hard. They are getting more done but working as hard or harder than before, with perhaps two caveats:

1. In the same way that heavy machinery saves humans from backbreaking labor, AI’s tireless clerical labor saves them from having to concentrate on a wealth of details and make fine distinctions in very limited time. So some very tiresome labor is now obsolete. It’s similar to a search engine filtering a pile of documents to find a handful containing a particular word or phrase.
2. The effort has shifted more to specification and planning and then review at a high level of results and output. For an essay, it’s less about checking for spelling mistakes and grammatical errors and more about the organization of the material and the flow of the argument or narrative. For software, it’s less about syntax and more about assessing architectural trade-offs and maintainability.

+ + +

“Data is the new oil. Like oil, data is valuable, but if unrefined it cannot really be used. It has to be changed into gas, plastic, chemicals, etc. to create a valuable entity that drives profitable activity. Data must be broken down and analyzed for it to have value.”
Clive Humby in 2006

h/t James Butcher

+ + +

“There is a growing mountain of research. But there is increased evidence that we are being bogged down today as specialization extends. The investigator is staggered by the findings and conclusions of thousands of other workers–conclusions which he cannot find time to grasp, much less to remember, as they appear. Yet specialization becomes increasingly necessary for progress, and the effort to bridge between disciplines is correspondingly superficial.”
Vannevar Bush in “As We May Think”

+ + +

“Our methods of transmitting and reviewing the results of research are generations old and now totally inadequate. […] Mendel’s concept of the laws of genetics was lost to the world for a generation because his publication did not reach the few who were capable of grasping and extending it; this sort of catastrophe is repeated all about us, as truly significant attainments are lost in the mass of the inconsequential.”
Vannevar Bush in “As We May Think“

Chalk Talk: Write Down Key Hypotheses

I encourage entrepreneurs to write down their key hypotheses because startups succeed through disciplined learning that builds on entrepreneurial intuition.
Instead of treating a business idea as a fixed plan, I view it as a set of testable assumptions that can be validated—or disproven—through conversations with prospects and early customers.

Every startup begins with uncertainty. The core questions are whether you are targeting the right customers, building the right features, and communicating benefits that customers actually value.

Writing these assumptions down forces founders to make them explicit instead of relying on vague optimism or diffuse beliefs.

Once documented, each hypothesis can be tested through customer interviews, prototypes, demonstrations, and early sales efforts.

I encourage entrepreneurs to document their expected outcomes or success criteria before investing significant time or money in new features. Entrepreneurs should define what evidence would confirm that an idea is working, what results would indicate it is failing, and when they will review progress.

By writing down milestones, expected outcomes, and stopping rules, founders reduce the risk of persisting with an ineffective strategy simply because they have become emotionally attached to it.

A critical aspect of this discovery process is asking prospects about past behavior instead of intentions, desires, or goals.

Prospects often share what they hope to do in the future, but intentions are a weaker level of evidence than past behavior. I recommend asking what they have already done to solve the problem, what alternatives they have tried, what resources they have committed, and what consequences they have experienced. Past actions reveal priorities far more reliably than future aspirations.

There is one useful exception. If a prospect can point to a written goal that is part of their quarterly or annual operating plan, that provides stronger evidence than a verbal statement. Especially if the customer’s organization has an active goal-setting and review process, the objective is considered strategically important, and management believes the goal is at risk.

A written objective that’s been reviewed and agreed upon suggests that the problem is not only widely visible but that resources are likely to be available to address it.

Entrepreneurship is more of a design and exploration effort than pure execution of a plan. The ability to focus and execute is necessary but not sufficient,
entrepreneurs must address problems or needs customers will pay to see solved.

Every initiative should include a written hypothesis, an expected range of outcomes, and a budget for time, money, and effort to substantiate or invalidate it. Customer conversations then provide evidence that either strengthens or weakens the hypothesis, allowing entrepreneurs to iterate intelligently instead of tinkering or making changes at random.

My experience has been that writing down key hypotheses improves decision-making because it forces you to reconcile observed outcomes with the results you predicted, rather than unconsciously adjusting your memory of your assumptions to match what happened.

Reviewing old hypotheses and assumptions can be embarrassing, but it  creates shared accountability, makes learning visible across the team, and helps entrepreneurs pivot intelligently based on evidence rather than fear or frustration.

This rigor enables founders to discover what customers truly value, reduces wasted effort, and increases the likelihood of finding at least a niche market interested enough in your offering to provide sustainable revenue.

Related Blog Posts

• Experiments vs. Commitments
• Jeff Allison: How to Drive Innovation and Meet Commitments
• Beston Jack Abrams on Curiosity, Experimentation and Action
• Organizing Your Experiment Log
• We Help You Design Experiments That Move Your Business Forward
• How To Run Experiments That Improve Your Business

Image Credit: (c) Theresa Shafer

This blog post was re-published on LinkedIn at https://www.linkedin.com/pulse/chalk-talk-write-down-key-hypotheses-sean-murphy-fw76c/

Bill Grosso on Startups after AI

Over the past 25 years, Bill Grosso has been the CEO of 4 separate software businesses (2 venture-backed startups, 2 bootstrapped consulting businesses) and has been a C-level executive at 4 other companies. During that time, the cloud emerged, the idea of analytics and machine-learning at scale emerged, open-source software completely altered both the development and deployment landscapes. And now, of course, Generative AI is changing everything yet again.

In this video, Bill draws on his experiences over the past quarter century and talk about how startups have changed, how startup customers have changed, and how we build software has changed, leading into a discussion of the role of startups in the emerging business ecosystem.

Artificial intelligence is transforming software development, but perhaps not in the way most people think.

In this talk, entrepreneur and software executive Bill Grosso argues that AI is best understood not as a revolutionary new form of intelligence, but as the latest in a long line of productivity-enhancing technologies. Like compilers or garbage collection before it, AI dramatically increases what skilled engineers can accomplish. The real disruption comes from how this changes the economics of startups.

Coase’s Theory of the Firm

Grosso begins with an idea from Nobel Prize-winning economist Ronald Coase’s The Nature of the Firm: companies exist because they reduce the cost of coordinating work. As organizations grow, however, they also become more risk-averse. Preserving existing products and revenue streams often takes priority over pursuing uncertain opportunities.

The Role of Startups

That’s where startups fit into the ecosystem.

Rather than viewing startups as future unicorns, Grosso suggests thinking of them as outsourced innovation engines. They tackle problems that large companies recognize but cannot easily solve within their own organizational structure. In many cases, acquisition—not an IPO—is the natural outcome.

AI changes this equation by making software development dramatically more productive. Teams that once required 20 to 30 engineers can now accomplish similar work with four to six highly skilled people using AI-assisted development. Smaller teams mean lower capital requirements, faster product development, and shorter startup lifecycles.

How AI Changes Startups

At the same time, AI enables large enterprises to build many internal tools themselves.

Previously, developing an experimental application might have required a dedicated innovation team. Today, a single engineer equipped with AI can often build a working prototype. Because coordinating with outside vendors remains expensive, many projects that once would have become startups will instead stay inside the enterprise.

This shift has important implications for entrepreneurs.

1. Software alone is becoming a weaker competitive advantage. AI makes it easier not only to build products but also to copy them. Traditional software moats are shrinking, meaning startups must compete on customer understanding, execution, relationships, and speed rather than code alone.
2. Startup funding will adapt and evolve. Faster product development puts competitive pressure on lengthy fundraising cycles. Early-stage investors will increasingly finance rapid experimentation, while larger follow-on rounds will focus on helping companies establish market leadership before competitors catch up.

Grosso also emphasizes that AI does not eliminate the need for disciplined engineering. His own teams rely on detailed specifications, acceptance criteria, automated testing, multiple AI models reviewing each other’s work, and experienced engineers providing architectural guidance. AI accelerates implementation, but good engineering practices remain essential.

Finding Startup Opportunities

Perhaps his most practical advice concerns finding startup opportunities. Instead of inventing technology first and searching for customers later, entrepreneurs should ask large companies a simple question:

“What important problem do you already understand but cannot solve because your organization gets in its own way?”

Those answers often reveal the best startup opportunities.

The age of AI doesn’t eliminate startups. Instead, it changes their purpose. Success will belong to founders who identify valuable customer problems, leverage AI to move faster than ever before, and build companies designed for rapid validation, sustainable growth, and—quite possibly—an earlier, more profitable exit.

Advice for Entrepreneurs

For aspiring entrepreneurs, Grosso offers practical advice. Be skeptical of startup mythology, especially advice from venture capitalists who naturally promote high-risk strategies. Focus on solving problems that large companies already understand but cannot address internally because of organizational constraints. Use AI extensively for market research to validate ideas before investing significant time. Finally, recognize that startup success increasingly depends less on writing code—which AI commoditizes—and more on identifying valuable problems, understanding customers, and executing rapidly before competitors catch up.

Key Takeaways

1. Advice for Entrepreneurs
   • Be skeptical of startup mythology
   • Question the incentives behind startup advice
   • Consider smaller, earlier exits instead of pursuing unicorn status
   • Think of entrepreneurship as an iterative process rather than a one-time bet
2. Other Observations
   • AI changes the economics of startups more than the purpose of startups.
   • Small, highly productive teams can build sophisticated software.
   • Success increasingly depends on identifying valuable customer problems rather than simply writing code.
   • The winners will combine AI-enabled execution with deep customer understanding and rapid market validation.

Other Resources

• View slides [PDF]
• Ronald Coase’s The Nature of the Firm [PDF]

Related Blog Posts

• Andrew Shindyapin: AI’s Impact on Software Development
• The AI Gold Rush: It Still Takes Teams
• The Hidden Cost of Replacing Junior Engineers with AI
• Mark Bennett on Using Claude Code for Application Development
• Mark Bennett: Using Claude Code in Teams
• AI in Action: Practical Automation by Alex Panait 
• Matt Trifiro on Lessons Learned using AI for Marketing
• Time to Market S01 E03: How Will AI like ChatGPT Impact B2B Founders?
• Alex Panait on Current Trends and Possible Futures for AI
• Game Data Pros Game Pros CEO Bill Grosso Discusses Future of Startups After AI

The Hidden Cost of
Replacing Junior Engineers with AI

Summary: Organizations need fresh talent and long-term investment to stay innovative. Apprentices challenge outdated assumptions, while experienced teams build reliable practices. AI can assist routine work, but lasting competitive advantage comes from learning organizations, shared knowledge, sound engineering judgment, and continuous capability development.

This conversation is a continuation of “The AI Gold Rush: It Still Takes Teams” where we explore other aspects of AI tools on engineering processes, organizational design, product development. Also the ill-conceived management trend of replacing junior engineers.

Edited Transcript

Norbert Korny: There’s the rush for AI to replace humans, and especially the juniors. Do you see the impact on the job market in Silicon Valley?

Sean Murphy: That’s absolutely going on. I think that’s ill-conceived for a couple of reasons. I don’t think it’s going to work that way. And if you don’t bring on any novices or apprentices, how do you get journeymen and masters 10 years from now? In other words, that’s a recipe for going out of business.

Jeff  Allison: Well, it’s the same thing that we did with outsourcing, right? And then all of a sudden, you know, you outsource everything, and then you go, okay, I need a power guy or I need a signal integrity guy, and you can’t find those guys. They’ve gone somewhere else.

Sean: You didn’t plant them five years ago to let them get better and work their way up.

Jeff: Right. You always assumed that someone else was going to do it, and there’s always going to be an unlimited supply of the talent you need. Sean, you’re right: bringing in young talent should be part of the culture of an organization if it wants to get better over time. We all get stuck in our ways, lapsing into “This is the way we’ve always done it, and it’s always worked.” Some bright kids come in, look around, and say, “Why are you doing it that way?” Which is great. It’s refreshing and gives you a lot of energy.  If you don’t bring in new talent, it may look great on the damn Excel spreadsheet for expenses, but fast-forward five years–maybe only three years–it’ll be a very different picture.

The numbers guys look at engineering today as a collection of costs and assume we can develop whatever we need when we need it. But view it as critical that they drive costs down. Well, that’s not necessarily true. Sometimes you have to make a huge investment early on to get a new technology under your belt so that you can reap the benefit in future generations of product. You can’t just add a significant new capability in one step. Trying to cut the cost right at the beginning doesn’t make any sense to me, but I’m not a CFO.

Newcomers bring fresh perspectives
that challenge established practice

Sean: I think you made a good point. Newcomers look at things with a fresh perspective  and they challenge established practice. Now, perhaps 80% of the time there are good reasons for traditional solutions, and they persist. But about 20% of the time people say, “How about that? The assumptions we made when we put this practice in place no longer hold, and we didn’t notice.” I don’t think that you can get an LLM to challenge you and say, “Hey, it seems like some of the assumptions implicit in your request or plan are incorrect, I think you need to check your premises or the basic assumptions behind your design decisions. Whereas a new person will go, “This doesn’t make any sense to me.” Now, that’s can be irritating, but sometimes it’s a grain of sand you can use to make a pearl.

Jeff: I was thinking about when we used to bring in new guys. They’d gone through the universities and were familiar with new tools and new methodologies. My worry here is that if these guys are learning these AI tools when they get their degrees, it may limit their ability to think outside the box and innovate. We need to be careful not to limit ourselves to what AI tells us. I think we need to watch out for that.

Sean: I agree, but the universities have a challenge there as well. The clock cycle on changes is running very fast right now. If I am teaching first-year calculus, differential equations, or another body of knowledge that’s been debugged over decades, the college-university system works great. But in this situation, where methods are being rapidly obsoleted, I think this argues more for an apprentice model. How do we attach juniors and others to the guys who are breaking the trail? How do we create an apprenticeship system and a community-of-practice model that encourages shared exploration, not just individual tinkering, to create a more effective learning dynamic?

Instead, the labs are selling a “tall thin designer” paradigm where one guy puts on a red cape and blue tights and does it all, encouraging the CFO to believe they can cut costs by letting go of a large chunk of their engineering or customer support staff. But the strange thing is that it’s the older, more experienced engineers who are getting better results. They can spot not only basic errors in generated code but also poor architectural choices the tools make. It’s led to the first example of reverse age discrimination I’ve seen in Silicon Valley in 40 years. Normally they tell the old engineer, “You’re too expensive, we can get a junior guy with twice the energy to do the job for half the salary.” Now we are replacing junior engineers and keeping the seniors

I’m not quite sure how to make the apprenticeship model work. We’ve had this long period of a kind of zero-interest-rate (ZIRP) paradigm, where we got used to things working a certain way. There are learning organizations.  NVIDIA seems like one, and I am sure there are many others. I think organizations oriented around learning will be able to digest these new technologies and move faster with them.

I don’t think we’re paying enough attention to how to build on initial breakthroughs. It’s less about the guy who gets the first 10x productivity bonus and more about reducing his insights and methods into an established practice. A learning organization is also a teaching organization, about fostering team-level and organizational practices that are more effective. To your point, Jeff, there’s a tension between encouraging people to compare notes and share what they have learned, encouraging wider adoption and additional refinements and improvements, and measuring just relative personal productivity. I don’t want to share what I’ve learned if I am facing a 20-30% layoff, because I’d be reducing my competitive advantage over my peers. Management is creating perverse incentives for shared exploration and shared learning.

Questions for Planning for a Paradigm Shift

Jeff:  Well, I think all of these issues are common to anything that’s early on a paradigm shift.  It’s easy if I’m starting from scratch, but if I’ve got a current roadmap, do I intercept what’s going on right now with this new technology, or do I look at something new, get my legs under that, and drive new skills back into other programs? And what’s the timing behind that? And who does that? Companies and organizations have to manage bringing on new talent, providing training, and developing new processes. Management asks questions like:

• What’s this do for my current roadmap and product portfolio?
• What does this mean for our longer-term strategy?
• What new  products should I start?
• How do I make this technology scalable over time?
• How does this affect my architecture?
• How does it affect my organizational structure?
• How should I evaluate our internal skills inventory?
• What mass training  do we need to offer to get people up to speed?

There’s just a whole list of things people need to kind of get their heads around. It’s not just about AI and how it’s going to make us magically more productive overnight. I think individuals may get an immediate boost, but figuring out how to make a product development organization more effective will take a while.

Sean: So Norbert, what do you see going on? You’re in a larger organization. How do you see this getting digested?

Norbert: Well, you need to have a deep understanding of systems and system design already,  because the tools don’t know how to build better systems.

Jeff:  That’s true. Tools don’t develop products, people develop products. They may enable you to develop more easily but you are still doing the engineering. Even with better tools you have to know what you’re doing.

Norbert: There’s a thin line between apprenticeship and blindly copying the response of the ChatGPT into the code. With this rush to embrace AI we have lost some of that and I don’t know if we are encouraging juniors to understand and make improvements on the ChatGPT output. Because at some point we just run out of seniors.

Sean: Early in the development of a new field or discipline information is fragmentary and contradictory. We are working on this hazy frontier, which we think is jagged  because of our ignorance, and we need to be organized in translating our exploration and tinkering into new methods that are reliable. You’ve got fragments of understanding you need to verify to untangle the contradictions, mark the blind alleys, and remix to find a reasonably reliable approach.

That’s not something that an LLM is going to help you with because there’s not a knowledge base you can train it on. If you want to teach somebody an established body of knowledge that’s been debugged, something like algebra, then I can use an LLM. But in a rapidly changing field where the frontier is mutating rapidly, there is not a lot of reliable information that’s written down. There are firsthand accounts, but many will contain errors and oversights despite the best efforts of the people involved. You cannot really Google for information that’s not well structured and expressed in common terminology. This is where apprenticeships and community-of-practice models are more effective, because much of the new information is shared only in conversations, emails, and informal briefings and presentations. Or it’s the result of watching someone else do something and gaining insight.

Jeff: So I give an agent a design challenge, and they write a piece of code for me. Part of the problems we used to have in engineering was the lack of reuse, right? So, someone would develop some really smart piece of code or smart circuitry, right? Then we would ask, “Why aren’t other people using it in their products? Why don’t they take what’s been debugged and put into production and use it to save time and effort?”

Well, part of the problem was that it wasn’t packaged in a way they could use. So they’d have to talk to the engineer and quiz them on some critical details they neglected to write down because they were trying to hit a ship date, not develop a reusable design element. And the developer may have worked on this design block two or three years ago and they are on a new project with new deadlines and they are not measured on helping out a different project. Sometimes new test cases have to be developed for a slightly different use case the block will now be applied to, and this involves more questions and perhaps some debugging of why it’s not working.

If the AI tools can not only generate the code but generate accurate documentation that matches it and test cases that enable reuse then it may have something to offer. Do you think AI will help with reuse?

Norbert:  It might.

Jeff: How do we have conversations around design reuse and standardization? How do we introduce methods for libraries and standard practices that align with our culture? How do we manage the security issues: both in terms of leakage of important intellectual property outside of the organization and risks introduced by subtle defects in the code that is generated?

Moving from Breakthrough to Beaten Path

Sean: I think the challenge we are talking about is moving from breakthrough to beaten path, from an artisanal proof of concept or prototype to an industrial-strength process. You have people exploring in many different directions, making unique discoveries and gaining distinct insights. But they are not necessarily compatible because they proceed from different perspectives and can diverge further over time.

What I’ve seen is that shared methods are typically negotiated outcomes. You need to get it documented, and then people have to look at that and say, “Okay, here are the strengths and weaknesses of this approach. Here are the strengths and weaknesses of that approach.” You can support multiple methods depending on the problem domain you’re working in.

Norbert: Right, but if AI can help facilitate that, it’s a good thing. The goal is better methods that yield better products faster, not smaller teams.

Jeff: Business guys are going to drive the businesses. They’re going to give engineering the amount of dollars they think engineering needs, which is rarely enough. So the engineers say, “Well, we can do this now, but it’s going to take longer. Or we won’t come out with full functionality in six months, we can only hit 60% at this level of funding.”

But those conversations always happen. What I’m trying to figure out is how does engineering become more efficient using this technology? My other concern is if you have something generating lots of code, how do you test all of it? Because you don’t want AI developing test programs for the code that it generated; that’s basic engineering practice. If someone develops a circuit block or a piece of code, you can have them write a basic unit test, but you need others also to write tests. So having AI test its own code doesn’t make sense.  So I don’t know how all of this will work out: a lot of code is being developed very quickly now. How do you develop a test process around that to verify that all these new modules work together? They may work on their own, but what about at the system level?

Norbert, say a build fails at 1 a.m. because someone checked in a piece of code that broke something, or you have all these conflicts. Now someone has to wake up, go through all the logs, look at the code, and untangle it.  Will AI help there, or make it worse, or how are people going to struggle with that?

AI is Tireless Diligent Intelligence

Sean: I think we should look at “AI” more as a tireless or diligent intelligence; I think in the break-the-build case, a simple set of rules could be applied consistently to catch basic mistakes. You attack the bottom 20-30% of the problem with AI instead of trying to achieve a breakthrough.  Ask it to do the bookkeeping and basic blocking and tackling.

My fear with the labor cost savings focus is that I have never worked in an engineering organization where the leader gathers everyone together and says, “This is a great team. We set very high standards, so I want to let you know that a year from now, only 60% of you will be here.”  Salespeople are used to that. When you bring in a new sales force, it’s kind of implicit that if they don’t make their numbers, they’re gone. But engineers are not normally managed that way because they’re working in teams, and you can’t always trace who contributed what. There are ways to assess contribution, but it’s more at the team level. I worry that this focus on cost reduction is going to prove extremely corrosive to leveraging the breakthroughs the pioneers make, using them to build on the early guys, and actually raising the standard of performance.

Jeff:  Don’t you think this will work it’s way out over the next nine to eighteen months? We’ve seen people come in with a new business model they claim will dramatically cut costs. At some point they realize it’s not going to work. It’s going to be painful, right?

Sean: I think it’s going to take a couple years and it’s going to be very bad in the interim.

Cost reduction thesis  has no basis in reality

Jeff: I’ve been through headcount reductions, not because we brought them on ourselves but because of things that happen  in the market or the national economy. It’s very difficult when you commit a program to develop a product that you promise to customers or a market, you need to size it correctly up front. When you don’t, you are in for a lot of pain through the whole development process. My concern is people have been advancing this cost reduction thesis that has no basis in reality.

Sean: There’s a category of algorithm that’s called a greedy algorithm: essentially, you look at the choices that are immediately available and choose the best one. Typically it’s the lowest cost or shortest distance. You don’t look ahead multiple moves or consider the consequences; you pick the best move and keep picking it at each turn. For certain problems, these algorithms are optimal. But in general, you cannot cost-cut your way to greatness, which seems to be what some firms are trying to do. We are also starting to realize the implications of a code base that’s been auto-generated, but I don’t think we’ve seen the full consequences of that yet.

Jeff: It’s going to take time to play out. I hope most companies are starting to put this technology in play internally on smaller projects or a subset of projects to get it ironed out. Then they’ll understand the cost structures and what’s needed to use AI in their mainstream product line.

Sean: I don’t know how much controlled experimentation is being done. It does not seem like many companies are used to acting as learning organizations. It’s certainly not the forte of most IT organizations, who are more comfortable with vendor-supplied cookbooks and predefined migrations. The challenge many companies now face is that the rate of change in the industry has increased by at least an order of magnitude, and their internal control and planning structures aren’t set up for that.

Norbert: As far as testing goes, you can make another AI model to test the codebase just to have a different point of view, but the bigger problem is that humans need to review that code.

Most of us are just not wired for that. We are wired for writing code, and now suddenly we need to read a lot of it.

Jeff: Yes, read it and understand it.

Hallucinations and Psychosis

Norbert: Another challenge is that a few hallucinations, sometimes more than a few, are mixed into all of this generated code. There is a strong temptation to skip climbing the mountain and get there in one jump, but it rarely works out. Hallucinations are intrinsic to LLMs.

I think LLM-induced psychosis is going to be recognized as a problem. Mr Dunning and Mr. Krueger will be excited because sometimes the LLM hallucinates exactly what you would expect, just to validate your worldview. You can get caught in a feedback cycle where your false beliefs are reinforced, further distorting your thinking..

Jeff: Do you think so? Aren’t companies going to create their own standards and their own learning modules? First, for security reasons; second, for proprietary information; and third, in the belief that we’ve created it so we know what’s in it.

Norbert Security concerns for malicious code will be a big part of this. Another risk that’s getting more attention: how do you protect your source code when you build using a mainstream AI model, how do you prevent them from stealing it? From what I’ve read, this has already happened multiple times. You just can’t rely on some AI provider.

Jeff: But if I were running a company and had to invest in creating my own LLM model that’s going to take some headcount and a serious investment, won’t it?

Edge or On Premises LLM Models

Norbert: You may not need to build your own model. You may be able to protect against data and IP loss by running a model on premises. We moved everything to the cloud on the theory that data was stored safely and redundantly for disaster recovery. But the AI firms may face different incentives than the SaaS vendors.

Jeff:  Reminds me how social media firms told us everything we posted remained our data and no one else could use it. Now we find out the AI companies were strip mining all that content plus all of the books and artwork for their generative models.

Sean: I wonder if there’s a pattern match to the web server code that’s now essentially all open source for Internet-facing stuff for a variety of reasons. I think the other thing that happened was the big providers had a good enough business, and it would have been obvious if they were stealing from their customers. So, for the most part, as far as we know, they didn’t steal our data or code: Amazon hasn’t launched 40 other applications that look a lot like those that were formerly hosted on AWS. But the incentive set facing the large model guys, I think, is very different, and you’re going to see more people running open-source models locally for a variety of reasons.

Jeff: But if they run numbers on what it would take to do this internally, I’m sure that’s more than just a handful of people, right?

Sean:  So the basic model can be a barn raising, a group collective project, and then you feed it your own data privately to extend it. And if I use an open source model and then extend it, then that may not cost a billion dollars. There’s a lot to play out here, and I think we have at least one round of bankruptcies or reorganizations here before 2030 or so.

Jeff: But it’s the same thing though. At the end of the day, this technology needs libraries and a whole stack of supporting tooling and data. We know that historically companies created software libraries, hybrid libraries, corporate standards, and everything else. I mean, that’s a huge investment, right?

Norbert: Basically, it’s libraries of the skills, and anyone can publish a skill, some may be malicious but they get included by default.

Jeff: Another thing, I want to ask: let’s take Company A and Company B who are competitors. They’re competitors, and they’re using the same toolset. Where’s the competitive edge? How does Company A become more innovative than Company B if they are both using the same libraries, the same tools. The outputs are, in my mind, probably going to be the same, not so different because it’s all based on the same thing. So where’s the innovation, and where do I get competitive advantage?

Norbert: Well, this is a great question because what’s stopping you from building something like Facebook when you have all the tooling at your hands?

Sean:  But it turns out that building the Facebook application is not the barrier to the next Facebook. It’s the fact that they’ve got a massive installed base.

Related Blog Posts

• The AI Gold Rush: It Still Takes Teams
• Expertise Acquisition Light Speed Barrier
• Organizing Your Experiment Log
• How To Run Experiments That Improve Your Business
• Jeff Allison How to Drive Innovation and Meet Commitments
• Experiments vs. Commitments
• Good Post-Mortem Questions Spark Learning
• Six Activities for Learning New Skills and Tools
• Impatience For Success Works Against Learning
• Buying a Map vs. Learning to Explore
• Entrepreneurs Need To See With Newcomer’s Eyes And Ask Stupid Questions

Image Credits:

• Overview Graphic generated with ChatGPT by Theresa Shafer.
• Antelope Canyon (c) Kevin Murphy, used with permission.

This post was republished on LinkedIn as https://www.linkedin.com/pulse/hidden-cost-replacing-junior-engineers-ai-sean-murphy-lig3c/

Expecting Too Much Too Soon

One mistake entrepreneurs make is expecting too much too soon.

In your mind, the product works. The business model is clear. Prospects immediately understand the demo, see the value, and money starts coming in.

But a useful rule of thumb is: “Nothing new ever works.”

Even getting prospects to pay attention requires iteration. Before someone agrees to watch a demo or evaluate your product, they first consume your message, positioning, or product description. If that does not resonate, the demo never happens.

And if getting attention is hard, getting people to pay money is ten to a hundred times harder.

Early-stage success requires target practice: experimentation, tinkering, careful listening, and steady improvement. You have to pay attention not only to what prospects say, but also to what they avoid saying.

A good way to protect yourself is to develop Plan B and Plan C before you start Plan A. If you talk to 20 people who fit your target customer profile and none react positively or offer constructive feedback, it is time to shift.

One of the hardest parts of entrepreneurship is that many things you feel confident will work generate no response at all. You have to learn to persevere in the face of a 90% failure rate.

Related Blog Posts

• Why is it so hard to get your first ten customers?
  • We like to use the analogy of a combination lock to describe the early startup learning process: there are four key aspects of your product that need to be in sync: Need, Impact, Customer Segment, and Message. Fundamentally, this startup combination lock requires four correct settings to open. To keep it open as your product and customer needs evolves they need to stay in sync.
• Ryan Waggoner: Maybe Startups Are So Hard Because We’re Doing Them Wrong
  • Investors naturally gravitate towards founders who either hit a billion dollars in a few years, or die trying (sometimes literally), and then investors and founders both are incentivized to craft this story that they only way to win is to win big, fast, and with all your chips on the line.
• New Market Exploration
  • New market exploration is one of the most difficult customer development challenges that entrepreneurs face. It requires strong interview and sense making skills. It also requires a commitment to carefully observe and document customer behavior–with a willingness not only to “go and see” but to have the imagination to ask “what if” to explore what could happen if you would remove one or more constraints imposed by the status quo. Entrepreneurs must see underneath the surface behavior of prospects and their self-reported problem descriptions to understand the cognitive model for a task or job to be done by your product. Developing a first prototype–or even what you think may be your first product–does not mean that the need for further interviews and imagination goes away, if anything it’s just the next basecamp for further exploration. New market exploration requires not only the confidence to form a vision of a solution but the willingness to be surprised and to admit where your mental map does not match the actual territory, integrating new “ground truth” to improve your product and business model.”
• Marshall McLuhan: How To See The Future
  • “To be a futurist, in pursuit of improving reality, is not to have your face continually turned upstream, waiting for the future to come. To improve reality is to see clearly where you are, and then wonder how to make that better.”
    Warren Ellis “How To See The Future” (Sep-7-2012)
  • “I have gradually come to appreciate that the really important predictions are about the present. What is happening right now, and what is its significance?” Robert Lucky
• Ten Mistakes Early Stage Bootstrappers Often Make
  • #7 Expecting Too Much Too Soon: Not Planning for “Target Practice”, Iteration, and Improvement

Sign up for an Office Hours Session
To Make Sense of What You Have Seen and Heard.

An early market can be like a combination lock, you may have discovered one or two of the keys to opening it, but it’s hard to assess partial progress. Exploring a market it’s like listening to 15-20 second snippets from songs you have never heard before, you have to make sense of what a number of folks are telling you: little sounds familiar and it’s hard work. If you are having trouble making sense of what you have learned from your initial early market conversations, please sign up for an no cost no obligation office hours session and we can help you make sense of what you have seen and heard. If you are preparing to scout a new market and would like a briefing for a descent into chaos, you are also welcome to sign up for an office hours session.

Image Credit: (c) Bondars (Licensed from 123RF Image ID : 33650284)

The post was republished on LinkedIn as https://www.linkedin.com/pulse/entrepreneurs-often-expect-too-much-soon-sean-murphy-t9bic/

July 4, Independence Day 2026

“History by apprising them of the past will enable them to judge of the future; it will avail them of the experience of other times and other nations; it will qualify them as judges of the actions and designs of men; it will enable them to know ambition under every disguise it may assume; and knowing it, to defeat its views. In every government on earth is some trace of human weakness, some germ of corruption and degeneracy, which cunning will discover, and wickedness insensibly open, cultivate, and improve. Every government degenerates when trusted to the rulers of the people alone. The people themselves therefore are its only safe depositories. And to render even them safe their minds must be improved to a certain degree.”  Thomas Jefferson in “Securing the Republic” (1784)

Jefferson quote suggested by DataRepublican in “Our 250th Anniversary” who highlights “Every government degenerates when trusted to the rulers of the people alone,” as a core responsibility for all citizens.

+ + +

John Adams: I have come to the conclusion that one useless man is called a disgrace; that two are called a law firm, and that three or more become a Congress!

While some quote collections ascribe this Adams, it’s actually from the movie “1776” screenplay by Peter Stone. I watched 1776 earlier this year, it’s available on YouTube, and was surprised how historically accurate it was. The concept of a “law firm” in Revolutionary times was not really how lawyers practiced, they were normally solo practitioners with a clerk and perhaps an understudy/apprentice (there were no written bar exams until mid-1800s). Even my grandfather, born in 1888, learned law by apprenticing himself to a practicing attorney. The ABA did not successfully lobby legislatures to require law degrees until 1930s.

+ + +