He’s also looking at the wrong layer.

The interesting thing isn’t Claude Code itself (although it’s a fun toy). It’s what you put around Claude Code.

I run a harness. Hooks fire on every tool call. Skills load when the work matches them. Memory persists across sessions. A separate evaluator agent grades the work after every deploy. The agent doesn’t get to mark its own homework.

This sounds like sysadmin trivia, but it really isn’t. It’s the whole pedagogy.

Eaton catalogs tasks. Twelve hundred PDFs sorted. Small apps built. Video chunked into timestamps. Useful. I do those things too. But task completion is table stakes.

Here’s what sits on top.

A hook blocks git push –force and suggests –force-with-lease. A hook stops a bare pip install outside a venv. A hook auto-runs the matching test file the second I save a source file. A hook nags me for acceptance criteria before I let an agent build new automation.

None of this came from a vendor. I built it. Each rule is a mistake I made once and refused to make twice.

That’s where the friction lives, and it’s where the learning happens: building the rails for your agent. Making the information that it produces more meaningful, and more meaningful to you.

The agent is building a memory of me; a modern elaboration of the digital twin concept. Sixty-some files: User profile, project state, feedback I gave it last month it would otherwise forget. Every time I tell it something non-obvious, it writes the note down. Every time it acts on stale memory, it gets corrected and updates the file.

This is a lab notebook. The lab notebook happens to be the agent’s working set.

Students should build one of these. Not download one. Build it.

The evaluator is the part most people skip.

When I deploy a script to my Mac Mini, the agent doesn’t tell me it worked. It runs a separate skill called /verify. /verify has its own criteria file. It SSHs in, reads logs, checks cron, parses real output. It reports PASS or FAIL.

The builder and the grader are different agents on purpose. Builders over-praise their own work. So do students. So do all of us. Split the two and the system gets honest.

This is the move Eaton’s piece is reaching for and doesn’t quite name. The friction isn’t gone. If you pay attention closely, you’ll notice it just moved. It lives in the evaluator now.

So what does this mean for a classroom?

First, we need to stop teaching prompts. Prompts are typing.

Instead, teach the harness. Teach hooks: what should be impossible to do by accident. Teach skills: what’s worth codifying as a reusable move. Teach memory: what’s worth remembering across sessions, what isn’t. Teach the evaluator: how do you know it actually worked.

A student who can answer those four questions can use any agent. They’ll understand more deeply how the thing itself works, and in doing so, begin to become an owner of it, rather than it owning them.

Eaton ends with a tentative hope. Maybe agentic AI can cut organizational friction so students focus on real learning. Maybe.

But I’d push it further. The harness is the learning. Building it is how you understand what an agent actually is. What it can be trusted with. Where it lies. Where it forgets. When to override it. You don’t get that from a prompt library. You get it from writing the hook that fires when the agent does the thing you told it not to.

We should be teaching the harness.

The post The Harness, Not the Agent first appeared on Jay Margalus.

Glasgow

The Glasgow ride is about flow and distance. It’s the route I reach for when I want to settle in, find a rhythm, and let the miles stack without constant interruption. The terrain rolls rather than punches, and the road pulls you through open farmland and small-town edges in a way that feels continuous. It’s mentally easy in the best sense, with enough variation to stay interesting, but predictable enough that you can think, listen, and just ride. This is my “clear the head” loop.

Big Hill

Big Hill does exactly what it promises. This is a short, honest test, with one main effort that defines the ride. I like it when I’m time constrained or when I want a reminder that strength still matters. There’s something satisfying about a route that doesn’t pretend to be subtle. You warm up, you climb, you deal with it, and then you’re done. It’s not a scenic ramble; it’s a benchmark.

Plank Road

Plank Road is another go-to. I love this one because it’s understated and local in the best way. It feels like a road that exists for the people who live there, not for people passing through. The riding is steady, the surroundings are quiet, and the experience is tactile with changes in pavement, light, and grade that reward paying attention. This is a route I ride when I want to feel in the county rather than moving across it.

Church (Parkway)

I recently had the pleasure of riding the Blue Ridge Parkway for the first time, and immediately fell in love with it. This ride has a split personality, which is exactly why I like it. You move from small roads into the Parkway, and suddenly the scale changes. The traffic disappears, the views open up, and the effort becomes long and measured. It’s not about speed; it’s about patience and cadence. The Parkway section gives the ride a sense of occasion, and the approach and return (lots of downhill!) keep it grounded.

Goshen Pass

Goshen Pass is the most “Virginia” of the five. Water, rock, trees, and a road that threads through it all. It’s a ride I save for days when I want to feel small in a good way. The pass itself demands respect, but the payoff is immersion—cool air, sound, and texture. This is less about training and more about being reminded why riding outside still matters.

None of these rides are exotic, and that’s the point. They’re repeatable, legible, and tied to place. Together, they’re my personal map of Rockbridge County—five ways of seeing the same landscape, depending on what kind of day I’m having.

The post Five Favorite Bike Rides in Rockbridge County first appeared on Jay Margalus.

Building Strange Things on Jetstream2

I spent time exploring Indiana University’s Jetstream2 and realized the real opportunity isn’t just “run a big model faster,” but to use the compute to prototype genuinely weird systems. The Plankton Network idea won the week — a distributed sensor-mesh that treats the ocean more like a living computation than a dataset. A close second: continuing some of the microscopy work that I did a while back, but pointing the device at other big (but small) problems that data can help with.

There’s this thing I do when encountering new tech where, instead of driving straight toward a solution to something, I play with it a bit first. That’s where we’re at with this — a toy eventually becomes an understanding, which leads to an insight.

Bowser: My Slow Internet Browser Moves Toward 0.2

My RSS reader-browser-hybrid (“Bowser”) continued evolving toward the thing I actually want: a browser that unclutters my brain instead of fracturing it. I worked on integrating something in the spirit of StumbleUpon — not the gamified version from 2007, but the quiet serendipity beneath it. A “give me something delightful but not algorithmically thirsty” button.

This might become a defining feature: a machine that nudges me sideways rather than down.

ourss Digest Re-Prompting

I’ve been tuning the prompt system for ourss so that the digests feel like something a human actually noticed rather than something that simply happened to pass through an API. The goal isn’t summarization; it’s texture. Giving people the sense of what was alive in their feed that week. Not personalization to me, but personalization to whoever’s using it. Harder than it sounds.

On Being an Artist Versus Running an Art Business

The long essay in the reading digest about making a living as an artist is a pretty good read. The author describes the work of earning a living with your art as strengthening a muscle — not just the creative muscle, but the operational one: pricing, shipping, talking to collectors, iterating on what resonates, learning Image-Market Fit by trial. The part I underlined:

“Business is simply a lens. It’s a set of knobs that help you understand how money is made and spent.”

That idea. The notion that every creative practice is just a different arrangement of the same knobs — is exactly how I teach entrepreneurship. It’s satisfying to see it articulated so directly.

Doctorow on Risk-Shifting and “Flexible Labor”

Cory Doctorow’s piece on so-called “flexible labor” was another standout. He reframes the gig economy not as flexibility but as risk displacement — capital off-loading its operational uncertainty onto workers, one cleverly-named contract clause at a time.

That analysis is relevant to almost everything I teach students about power, systems, and what actually happens underneath business models. The more I read Doctorow, the more convinced I am that entrepreneurship students need a required systems-literacy course — a “here is how power is engineered into the infrastructure of markets.”

Space Apps, Badges, and the Ongoing Fourfold Push

I spent a little time this week refining a Cardano proposal with my guys at Fourfold. Every time I return to the THOTCON 0xD badge work, I’m reminded how much latent potential physical computing still has. We’ve barely scratched the surface of using hardware as onboarding — not gimmicks, but real worlds you can touch.

This week’s idea: a badge-to-blockchain bridge that treats the machine like a physical wallet. It’s strange enough that it might work.

Students, Schedules, and the Quiet Parts of Teaching

A lot of my week was spent smoothing things out for students… scheduling mentor visits, rewriting instructions so they make sense at 8:05 a.m., and keeping projects moving without killing the parts of the class that make it fun. We’re deep in the season where students are stretched thin and the work becomes less about content and more about creating enough clarity for them to move forward confidently.

A Final Thought: Adjacent Familiarity

The “adjacent familiar” concept from the art essay is a thing we teach in design — the idea that once someone loves a certain form, they want variations that sit just to the side of what they already know. It’s true of art, of games, of hardware, of teaching, and of programs like the one we’re building at W&L.

Everything new we’re building — Bowser, ourss, the AI models, the Cardano badge system — is really just adjacent familiar to something I’ve been circling for years: tools that restore human agency by stripping away unnecessary complexity.

A good reminder for the week ahead.

>> Oh, and check out my student MG’s new book on Amazon.

What a fantastically wonderful time this all is.

The post Weekly Notes — November 17, 2025 first appeared on Jay Margalus.

A lot of the work I’ve done this week has been about untangling short-term obligations from long-term voyages. Sometimes I confuse the two. Maybe my students do, too.

Meta’s fraud files (and what they say about design)

Doctorow’s dive into Meta’s internal memos reads like a case study in what happens when a platform decides the externalities are someone else’s problem. Fifteen billion scam ads a day. A quota for how much fraud the fraud team is allowed to prevent. My students ask me why I push them toward building actual things—physical prototypes, pieces of software they can explain, not pitch decks. This is why. If the substrate is compromised, the only stable footing you can give yourself is craft.

Relatedly:

I re-read weirder, older ubiquitous computing papers this week—Mark Weiser et al. on “interface agents” back when that meant something closer to a helpful ghost in your machine than a graph transformer wedged into a website sidebar. It pairs strangely well with Amazon’s legal theatrics over Perplexity’s shopping agent. Agents collapse the space between intention and action; platforms depend on that space staying large. That tension feels like it’s about to define a decade.

Hardware, small and joyful

The new USB-C Arduboy showed up in my feed: credit-card sized, 128×64 OLED, multiplayer over a cable like it’s 1992. I love that this exists. There’s something deeply reassuring about people continuing to build tiny game consoles with honest constraints while the rest of tech culture sprints toward frictionlessness. I’m sketching the next phase of the handheld dial-driven AI toy we’re prototyping—trying to understand how physical metaphors can make generative systems more legible. Sometimes the smallest boards make the biggest ideas possible.

Privacy as a design material

Another thread: Doctorow again, on why the internet—despite having world-class encryption running underneath everything—still feels like the worst privacy landscape we’ve ever built. The short version: policy, not math. I keep thinking about how this affects the way I teach entrepreneurship. Students want to build tools that help people. I want them to understand the architecture they’re building into. If you treat surveillance as a default, you end up designing for compliance instead of possibility.

One last link

This week’s reading included a LessWrong guide for whistleblowers in the U.S. government. Dense, serious, oddly hopeful. The existence of such a guide suggests that the long view is still available to us, if we’re willing to sail in that direction.

Thanks for reading. If any of this sparks something for you—about making, teaching, or the shape of the internet we’re handing off to the next generation—I’d love to hear it.

—Jay

The post Weekly Notes — 2025-11-10 first appeared on Jay Margalus.

The post ourss – An Opinionated RSS Reader first appeared on Jay Margalus.

The post OmniFocus Google Chrome Plugin first appeared on Jay Margalus.

After almost two decades working in hardware, I get it. Bits travel at the speed of a git push; atoms obey lead times, torque specs, and the occasional law of thermodynamics. Spinning a new PCB? Weeks (at best!). Shepherding a Class II medical device through the FDA? Let’s talk in quarters. Meanwhile, my software-only friends can refactor production over lunch. That cadence mismatch breeds all sorts of optimistic timelines.

When I’ve flagged this recently, the replies drift into hand-waving. Armchair architects sketching skyscrapers on napkins and assuming the steel crew will “figure it out.” But the fact is that the world of matter is less forgiving. Steel still yields at 36 ksi, silicon still needs a mask set, and the supply chain still runs on boats and trucks.

None of this means AI isn’t useful. It absolutely is. Vision systems catch bad solder joints before they ship; predictive maintenance saves factories real money. But headlines claiming the singularity is six years out ignore the slog of validation, liability, and, well, physics. Before you hand over your clinic to a black-box model, ask an actual doctor what they think. The raised eyebrow you get is the story.

Here’s the point: AI will augment physical work, but it won’t magic away the beautiful, stubborn complexity of the real world. Progress in hardware moves at the cadence of freight, not fiber. Any roadmap that forgets that rhythm is just a slide deck waiting to be walked back.

The post AI Meets Atoms – A Hardware Reality Check first appeared on Jay Margalus.

Key Highlights:

• Intrapreneurship and Social Entrepreneurship: Wendy discusses her passion for social entrepreneurship and how her upbringing surrounded by entrepreneurs naturally led her to intrapreneurship. She emphasizes the importance of this role in driving innovation within institutions and preparing students for rewarding careers.
• Building the Simon Center: Wendy shares her experience of building the Simon Center from the ground up, highlighting the significance of having seed funding and supportive leadership. She reflects on the challenges and motivations behind establishing the center and how it has evolved over time.
• Innovation vs. Entrepreneurship: The episode delves into the differences between innovation and entrepreneurship centers, with Wendy explaining her decision to prioritize innovation in the center’s name to make it more interdisciplinary and accessible, especially in a liberal arts environment.
• Challenges in a Liberal Arts Institution: Wendy candidly discusses the initial skepticism she faced in a liberal arts institution and the strategies she employed to gain acceptance and foster collaboration across disciplines.
• Student and Faculty Engagement: Wendy explores the balance between focusing on students and faculty, emphasizing the importance of student-led initiatives and the role of faculty in reaching a broader student audience.
• Mentorship and Community Building: The episode highlights the mentorship approach at the Simon Center, where Wendy curates mentor matches based on industry and expertise, and the importance of building a sustainable ecosystem that balances technology with human relationships.
• Future Trends and Challenges: Wendy shares her thoughts on the future of higher education and entrepreneurship centers, discussing the impact of changing demographics and the potential of AI in shaping the educational landscape.

  Join us for a thought-provoking conversation with Wendy Bolger as she shares her vision for fostering innovation and entrepreneurship in higher education and beyond.

View on Zencastr

The post TSW* S1E6 – Wendy Bolger – Loyola University Maryland first appeared on Jay Margalus.

From Conception to Prototype

The journey from an idea’s conception to a working prototype is often fraught with challenges. Traditionally, this process might involve a suite of tools—each with its own learning curve and idiosyncrasies. However, Figma consolidates these steps into a single, cohesive platform. This streamlining is crucial in an educational setting, where time is limited and the focus should be on nurturing creativity and entrepreneurial thinking, not on navigating complex software ecosystems.

Brainstorming with the IDEO Butterfly Technique

One of the first steps in the entrepreneurial process is brainstorming, and the butterfly test (invented by Bill Moggridge and highlighted in the book Change by Design) is a favorite in my classes. This method encourages expansive thinking, allowing students to explore a wide array of ideas before narrowing them down. Tim Brown describes it in Change by Design:

“”Invented by Bill Moggridge, design thinker extraordinaire and one of the pioneers of Silicon Valley design, the butterfly test is a thoroughly unscientific but amazingly effective process for extracting a few key insights from a mass of data. Let’s imagine that by the end of a deep research phase, numerous brainstorming sessions, and endless prototypes, an entire wall of the project room has been covered with promising ideas. Each participant is then given a small number of small Post-it “ballots” to attach to the ideas they think should move forward. Members of the team flutter about the room inspecting the tableau of ideas, and before long it is clear which ones have attracted the most “butterflies.” Of course, all kinds of issues come into play, including politics and personalities, but that is what reaching consensus is all about. Give and take. Compromise and creative combination. All these and more play a part in reaching the end result. The process is not about democracy, it is about maximizing the capacities of teams to converge on the best solutions. It’s chaotic, but it works surprisingly well and can be adapted to the peculiarities of many organizations.””

Figma facilitates this process beautifully with its collaborative features. Students can simultaneously contribute to a shared canvas, sketching out ideas and building on each other’s concepts in real-time. One great tool that students can use during this phase is Jambot.

Jambot, an AI-powered assistant integrated within FigJam (a part of Figma), offers a range of functionalities that can significantly enhance brainstorming sessions like the butterfly test. This AI tool leverages OpenAI’s technology to assist users in generating ideas, organizing thoughts, and refining concepts. Imagine a student group has 50 post-it notes of business ideas, and then they’re able to connect seemingly disparate post-it thoughts together and ask an AI to brainstorm business ideas for them. Other ways to use Jambot:

• Idea Generation: Jambot’s “AI Idea Engine” can be particularly useful during the expansive phase of the butterfly test, where the goal is to generate a wide array of ideas. By inputting a concept or prompt, Jambot can provide a diverse set of suggestions, helping participants explore various possibilities and directions.
• Visualization and Organization: During the convergent phase of the butterfly test, where ideas are narrowed down and organized, Jambot can help visualize concepts using diagrams and sketches. This aids in clearer communication and helps participants see connections between different ideas, facilitating the selection of the most promising ones.
• Summarization and Refinement: After the brainstorming session, Jambot can quickly summarize the sticky notes and ideas generated. This feature is useful for wrapping up the session and ensuring that key insights are captured and can be further refined or developed

Ultimately, this fosters a dynamic environment where creativity can flourish, and ideas can evolve organically.

Wireframing: The Blueprint of Ideas (Low-Fidelity Prototyping)

Once the brainstorming phase yields a viable concept, the next step is wireframing. Wireframes serve as the blueprint for any application, outlining its structure and functionality. Figma excels here, offering a range of tools and templates that allow students to create detailed wireframes with ease. Its intuitive interface means that students can focus on the design and functionality of their application, rather than getting bogged down by technical hurdles. Moreover, Figma’s ability to create interactive prototypes from these wireframes provides students with immediate feedback, enabling them to iterate quickly and efficiently by developing a quick and dirty prototype to show to stakeholders and — sometimes, even — users. The quickness of this process also means that students aren’t as married to their product idea yet — if it only took 15 minutes to make, no big deal if you need to throw it away.

Building High-Fidelity Prototypes

The final stage in the prototyping process is the creation of high-fidelity prototypes. These prototypes are crucial as they provide a realistic representation of the final product, complete with visual design and interactive elements. Figma’s robust design tools and extensive library of components make it possible for students to bring their ideas to life with precision and creativity. The ability to test these prototypes in a simulated environment (on, say, an iPad or iPhone) allows students to refine their designs and ensure they meet user needs before moving to production.

Figma even has templates for various devices — computers, phones, tablets, and so on — that allow students to create polished-looking experiences. More than that, Figma has a library of demo applications that students can take and retool for their own needs.

A Unified Tool for Diverse Needs

What makes Figma truly remarkable is its ability to serve as a one-stop-shop for the entire design process. This is particularly beneficial in an educational context, where the goal is to equip students with practical skills they can apply in real-world scenarios. By mastering Figma, students gain proficiency in a tool that is widely used in the industry, enhancing their employability and preparing them for the challenges of the entrepreneurial world.

But Figma is more than just a design tool; it’s a catalyst for innovation and creativity in the classroom. By simplifying the design process and fostering collaboration, it empowers students to transform their ideas into reality. As an educator, there’s nothing more rewarding than witnessing that transformation, and Figma plays a pivotal role in making it happen.

If you’re looking to get started with Figma, check out this great four part series they’ve put out.

The post The Power of Figma in Entrepreneurship Education first appeared on Jay Margalus.

TSW* welcomes Ellen Schmidt-Devlin, co-founder and Executive Director of the Sports Product Management Program at the University of Oregon. Ellen’s illustrious career spans 27 years at Nike, where she worked closely with legendary coach Bill Bowerman and gained invaluable experience in product development, merchandising, and international business management.

Ellen’s journey is marked by her passion for advancing women’s sports and her commitment to sustainability and innovation in the sports product industry. In our conversation, we delve into her experiences at Nike, the challenges and triumphs she faced, and her pioneering efforts in establishing the SPM program, which equips students with the skills needed to excel in the sports product sector.

We also explore her documentary, “We Grew Wings,” which celebrates the history and achievements of the University of Oregon’s women’s track and field teams, highlighting the impact of Title IX and the ongoing fight for gender equality in sports.

Whether you’re an aspiring athlete, an industry professional, or simply interested in sports and innovation, this episode is packed with insights and inspiration from Ellen’s remarkable career and dedication to nurturing the next generation of leaders in sports product management.

Tune in to gain valuable perspectives and be inspired by Ellen’s journey and her relentless pursuit of excellence in the sports industry.

View on Zencastr

The post TSW* Podcast ep. with Ellen Schmidt-Devlin first appeared on Jay Margalus.