There is a huge amount of administrative work that every parent ends up doing, and it only gets more complicated if you have more than one child or more than one school or sports situation.

I find this interesting because it’s an example of AI used to shift the power dynamics of a shitty system in the favor of a person, instead of the other way around.

We all have to exist within and use systems that are not optimized for us, and that we can’t change. These agents are one tool that we can use to reface complex and cruel systems into something that works for us, in the ways that we need.

If I were a VC (I am not!) I would propose a category here and call it consumer-empowering enterprise software or something along those lines; and look for the markets where:

1. the consumer must use the software but the provider has no incentive to care about their experience, with high switching costs (you won’t change to another school because you don’t like the way they share announcements, you won’t quit your job because HR uses a bad software system),
2. this often happens in ecosystems where the full experience isn’t owned by any single entity or platform (cough, healthcare, cough), and
3. where the core need is information management and consolidation, and there are real consequences for failing to perform and
4. where the system benefits from friction, or at minimum doesn’t care about you experiencing it (insurance claims, appeals, warranty support, etc.) and
5. where the ability to have an agent ultimately take action for you would be an absolute cherry on the sundae cake and provide a ton of value. (ie, handle the customer support bot to get the warranty form, fill in the form, follow up after six weeks when there’s been no response, write the note to the credit card company before the deadline, etc.)

Given that framework, there are lots of opportunity spaces to build in here, for yourself, or as a tool for others:

• healthcare coordination — if you have even mildly interesting health considerations, coordinating information across providers, making sure follow-ups happen, and meds are organized, that questions are answered
• Insurance and individual benefits management — help optimize your plan selections and make sure you track, file, and use your benefits, and get the care you’re entitled to
• Personal tax information management — if your taxes are even mildly interesting, record and file the receipts and docs, and highlight best strategies to prep the package for your accountant (or more)!

And so on with:

• Home ownership
• Estate settlement
• Vehicle acquisition and maintenance
• Eldercare
• Any project where you might hire a general contractor
• Customer support, and getting service your warranties entitle you to
• Managing parking tickets and other straightforward legal matters
• Group travel planning and on the ground coordination

So look for the spaces with severe fragmentation, high coordination cost, long-lived state and needs to follow up repeatedly, lots of documents/messages/information to manage, and enough economic or emotional value and high enough consequences for mistakes that just collecting that information is hugely beneficial.

It’s also worth saying out loud that the people who can build these agents themselves today are privileged enough to have knowledge and access to do so; we should focus on making the tools open and accessible for everyone who would benefit from taking some of the power back from these systems.

PS. If you want to play with building this sort of thing yourself, please be very thoughtful about the security and risks, use a system (like nanoclaw) that helps manage those things, and consider if a local LLM is necessary to keep information private!

Originally published on Medium.

I recently looked at my texts to myself and realized it seems like a reasonable set of random numbers. Hello, hotel-based entropy*!

What can you do with a very very very slow random number generator? You could spend a lifetime generating enough random digits to encrypt a document (you can seed a pseudorandom number generator, of course, but what’s the fun in that?).

* Okay, this is a really bad idea. For one thing, I only have seven numbers right now, and that’s not even enough to run against any of the dieharder tests. For another, they aren’t at all random, as there’s a bias in the number of floors and number of rooms in various hotels. Also, 5160 was in Vegas, and so even though I was told to go to the “51st floor” when I got off the elevator and looked out the window I was on the third floor. Vegas is weird. If you do actually want some real random numbers, check out the NIST Randomness Beacon. Not my hotel rooms.

** Okay, here are the actual numbers:

11085
12321
707
613
5160
504
pack glasses
1534

So what are people going to do with this? It’s easy to imagine high-quality voice output from Amazon Echo or Siri, or even scary applications, like scam phone calls that sound like they are coming from your own friends or family members (explored here).

And that brings us to today’s bad idea. Let’s build an app that allows you to have a conversation with yourself… in your own voice. You can share all of your half-baked ideas and arguments, and have your own voice cheering you on and repeating things back to you!

Every Friday, I’m going to share a ridiculously bad idea. This should be very easy for me (I have a ton of them), it’ll keep me posting on a regular cadence, and it should also motivate me to write more, since I don’t just want this blog full of bad ideas.

Today’s bad idea is for storage of immutable historical data. When you have data you need to put in cold storage that won’t ever change, why not build a giant XY carving robot, install it against a blank wall in your office lobby, and have the data chiseled into the wall? You can even come up with a design scheme that makes it beautiful. You solve your storage and interior decoration problems at the same time!

Need to retrieve the data? Just take a photo, and decode it. The idea of storing digital data on old media isn’t new. I’ve just yet to see it on a grand scale.

I had to look up how to get OS X Sierra (10.12) to remember my SSH keys. This behavior has changed to mimic standard OpenSSH behavior, which is a good thing!

First, add the key to your OS X keychain:

ssh-add -k [key_file]

Then, when you start a shell, run this:

ssh-add -A

I added it to my .bash_profile and it’s working quite nicely.

I’d like this site to help you figure out who I am and what I do, share what I’m working on now, let people get in touch, and be a place for me to share some personal thoughts and works-in-progress.

If you have suggestions, I’d love to hear them.

I’m excited that my short book, Data Driven: Creating a Data Culture, co-authored with DJ Patil, is out in the world!

We talk about processes and qualities of strong data teams and how to design for these cultural practices in an organization.

The book is available for free on O’Reilly’s site, and soon on Amazon. I hope you enjoy it.

Something fun to start off the new year — I found myself in Google Street View!

(A quick search of tumblr posts with streetview in the URL leads to a lot of fun, related stuff.)

Posted: February 14, 2014 | Author: Hilary Mason | Filed under: blog | Tags: data, food | Leave a comment »

I spent a few minutes this week putting together a quick script to pull data from the Locu API. Locu has done the hard work of gathering and parsing menus from around the US and has a lot of interesting data (and a good data team).

The API is easy to query by menu item (like “cheeseburger”, my favorite) and by running my little script I quickly had data for the prices of cheeseburgers in my set of zip codes (the 100 most populated metro areas in the US).

I’m a big fan of Pete Warden’s OpenHeatMap tool for making quick map visualizations, and was able to come up with the following:

The blue map is the average price of a cheeseburger by zip, with the red one showing the average price of pizza. The most expensive average cheeseburger can be found in Santa Clara, CA, ironically the city currently hosting the Strata data science conference this week. Have fun with those $18 cheeseburgers, colleagues!

You can also see some fun words in the pizza topping options:

In this plot, the x-axis is roughly geographic (ordered by zip code) and the y-axis is in order of popularity, with pepperoni being the most popular common pizza topping, and anchovies among the least.

This is just a quick look at some data, but hopefully it’ll encourage you to play with your food (data)!

Posted: December 18, 2013 | Author: Hilary Mason | Filed under: blog | Tags: email, hack, twitter | 3 Comments »

It turns out that it’s pretty easy to co-opt Twitter’s Lead Generation card for anything where you want to gather a bunch of e-mail addresses from your Twitter community. I was looking for people willing to alpha test a little side project of mine, and it worked great and didn’t cost anything.

The tweet itself:

Love tech discussion but looking for a better community? Help me beta test a side project! https://t.co/H3DYjbCy19

— Hilary Mason (@hmason) December 12, 2013

I created it pretty easily:

1. First, go to ads.twitter.com, log in, and go to “creatives”, then “cards”.
2. Click “Create Lead Generation Card”. It’s a big blue button.
3. You can include a title and a short description. Curiously, you can also include a 600px by 150px image. This seems like an opportunity to say a bit more about what you’re doing.
4. You do have to set up a privacy policy URL. I used a simple google doc.
5. You also need to configure a fallback URL, which is where people will go if they don’t have a Twitter client capable of the one-click signup. I used a Google form, which let people give me their e-mail addresses directly.

And that’s it! Tweet enthusiastically, then wait patiently, because if you don’t integrate your Twitter card with your CRM, you have to wait ~24 hours for the download link to appear in the Twitter cards manager. The resulting CSV looks like this:

Timestamp,User id,Name,Twitter handle,Email
2013-12-12T23:36:05,774485611,Robots Rule,RobotzRule,h+robots@bit.ly

A little bit of awk later and I had a list of e-mails ready to go. I ended up getting 49 responses through the Google form and 197 through the Twitter card. It was easy and I’ll definitely do this next time I need to collect people’s e-mail addresses for a project.