Package products and services for the fast-growing personal AI channel. Mountain bike crash tester. All of these words are written by a human.
Boston, MA
Joined October 2007
- Tweets1.7K
- Following797
- Followers944
- Likes988
Pinned Tweet
Read this article to learn about the long and colorful history of AI doom. Look for doomer jargon in bold font. Find out where we are in the doom loop. Understand ways that AI can hurt you now.
AI must work. Anthropic is bringing in $6.6M revenue per employee. Net revenue retention is 900%, meaning that the more people use the product, the more they want to use it, to the point where they buy 10X as much. "From September 30, 2025 to September 30, 2026, we have grown customers with $100,000 or more of revenue on a TTM basis from more than 950 to more than 11,900." These numbers are so unusual in the history of S1 filings that they might justify saying "this time it's different". Also the S1 is 400 pages so you probably want AI to help you read it.
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO.
Read more: anthropic.com/news/s1-sec
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
Speaking at @AITinkerers Boston! Presenting "OAuth for AI Clients" boston.aitinkerers.org/p/ai-…
AI Tinkerers made this heroic image and I deserve it because this topic is important.
On Monday in Cambridge I will show how your users can login and connect from any AI.
Register now to secure your spot:
boston.aitinkerers.org/p/ai-…
Join us for a demo-first evening focused on practical AI applications in security. We'll hear from William Haver on runtime governance for AI agents, Connor Whelan on building decision models for cybersecurity with Copley, Craig Chamberlain on developing a patent-backed security AI portfolio, and Andy Singleton on using OAuth as a universal gateway for AI tools. This is a fantastic opportunity to see live builds and learn from builders tackling critical security challenges.
ABOUT AI TINKERERS
@AITinkerers is a global community for hands-on AI builders. I thank and @natea for organizing an insightful and engaged meetup in Boston.
ABOUT BOTCONX
Botconx is a studio that connects you to personal AI users. Call us when you want to simplify and accelerate adoption of your services. This is our first showing of the Botconx platform. Learn more at botconx.co
Wins today's award for making lemonade from lemons
If you wanted to know all the jobs that AI will produce, look no further than cybersecurity. This is going to become one of the biggest investment areas for enterprises going forward.
And even as the work becomes more agentic and autonomous you will need more talent to point the agents at the most critical work, review findings, patch systems, respond to incidents, and more.
Lots of opportunity right now for the systems that protect data and enterprises, as well as the professionals that are deploying them and protecting organizations.
It’s going to only get more wild out there.
More on the death of software product management: Stephan Smith writes "one of the things that I’ve been thinking a lot of is that we might be having to rethink some of the basic premises of sas and it may be that software that’s written for one organization is actually the dash and you should own IP and not software written for 1000 clients "
---------
I agree. In the past it was important to do rigorous product management and only include features that multiple customers can use. That keeps the product maintainable and high quality. Now we have AI to manage the complexity of just writing and rewriting a different version for each user, and maybe that is the right way to offer it.
The AI companies are pushing us toward a stack where the only thing that we maintain is a bare connector. On top of that are skills that the users write and share. On top of that is "generative UI" (like ChatGPT 6) that the AI makes and customizes for each user. In that world, you don't maintain a customized version of an app. You just ask the AI to show a custom UI as needed. This is the type of app that the AI will build on top of our new Botconnect primitives.
We have budget limits and approvals to contain the “insider risk” of a human employee that makes mistakes. The same controls will work for an unreliable AI. This applies to any trading or expenditure and not just enterprise management.
Let’s do it. Botconx builds plugins, and tests, places and maintains them to reach users in all of the major chat, desktop, and bot platforms. HTTPS://botconx.co
I literally can't think of a better risk/reward bet right now than building a ChatGPT plugin and getting in front of their 1.2 billion weekly users.
As of yesterday, ChatGPT recommends plugins in the middle of conversations. Millions of questions get asked every day with no plugin to recommend. Each one is an empty slot.
Facebook apps in 2007. iPhone apps in 2008. TikTok in 2019. Every big platform has a few months where distribution is free, and most people miss it.
I saw this firsthand when I lived in SF building venture backed companies. They are sacred times!
If I were starting this week:
1. Find a tool or workflow people already Google for or do by hand every week (PDF to Excel, invoice generator, weekly client reports)
2. Build it as a ChatGPT plugin
3. Write the description using the exact phrases people type into ChatGPT
4. Ship 5 of them and see which ones get picked up
5. Double down on the one with the most traction
In 12 months these slots will be crowded but right now they're pretty empty.
Most people will realize ChatGPT plugins were the biggest free distribution channel of 2026 sometime in 2027.
The rise of Instinct and Muse shows that we are in a rapid transition from web apps with a human browser UI, to a UI where we just talk to our AI and the AI handles the apps. We are producing more services and more software with more features, and it's too much for human users to absorb. The AI interface is much simpler for the human. The AI can absorb complexity and use any number of products and features with new techniques like progressive skill discovery.
This is the ONE growing channel for new products and services. The usage of Web apps and phone apps has flatlined. Plan your product launches accordingly.
FRIEND COMMENT: There is currently no repsonsible way to let it at information for which you have a responsibility of care (such as company data).
ANDY: I am building "connectors" which will attach to these bots with one MCP or skill, and provide managed access, along with a policy editor. You can package a company's systems that way, or a part of your life, such as wealth management.
FRIEND COMMENT: That's a good product! Letting your employees BYOB (that's Bring Your Own Bot) in a way that's safe and responsible is the obvious future.
Why this might be important: PE investors are starving for exits on $3T worth of assets. One of the main reasons that PE deals don't IPO is that they are too small.T he "market structure" of public markets has changed so that companies with market caps of less than $1B do not get attention or trading liquidity. That is bad. It means that when someone sells, your stock price goes down, and you look bad. Companies avoid getting into this situation with NMS. Crypto markets often have very active trading for small caps down to $10M.
After reading the SEC order, it looks like this AMM exemption is not going to be useful for capital markets. It restricts the AMM to trading 2.5% of the average daily volume in the bigger NMS markets. So, you can't use it to add enough AMM liquidity to improve a security offering. As soon as the trading gets interesting, the token will hit the volume limit and trading will stop. Then if you want to sell your tokenized stock you will have to redeem it into real stock and sell it somewhere else, or find a buyer without using an AMM.
One theory for why small caps have died on normal stock markets (NMS) is that the SEC changed the rules for marketmaker spreads. They forced the spreads to be smaller. This makes trading cheaper for large cap assets that go into ETFs - the core product of American markets. However, it's bad for small caps because they lose their marketmakers and nobody trades them. I am not sure if that is the primary mechanism, but I am sure that the SEC has made a lot of small changes over 30 years that make US public stock markets better for large caps and worse for small caps. So, we have world-beating large cap and ETF markets, and the small caps are private.
When I tried this in 1993 I built wooden racks and installed 100 single-board computers. I also designed my own stack language (GPQuick) to run genetic programming efficiently. The random genetic operations for what is not called autoresearch were MUCH less efficient than our smart LLMs, and we had many orders of magnitude less compute and data.
Today everyone is talking about Recursive Self-Improvement (RSI). In 1987, when compute was 100,000,000 x more expensive, I published the 1st concrete RSI algorithms. Now compute is cheap, and RSI is driving the future of both software and physical AI. See: RSI since 1987 people.idsia.ch/~juergen/rec… (Technical Note IDSIA-9-26)
Also covered: RSI with self-modifying policies since 1994, gradient descent-based RSI in neural networks since 1992, asymptotically optimal RSI for curriculum learning since 2002, mathematically optimal RSI through the self-referential Gödel Machine since 2003, RSI combined with artificial curiosity and intrinsic motivation since 1990/1997, recent work on RSI since 2020.
Software-based RSI has become practical. Full RSI, however, will require not just self-improving software but self-improving hardware in the physical world.
As of 2026, companies talking about RSI include Anthropic, OpenAI, Sakana AI, SpaceX, Ricursive, Recursive Superintelligence, Inherent …
This behavior will pop up if you ask your agent to spend or invest money. It might decide that you really need $1M worth of Brooklyn Bridge shares, or a flashy watch. You need to have a different process that checks and approves expenditures, so that it can reject these overenthusiastic actions. However, if that wallet or policy approver is running on the same computer as your AI, the AI will try to hack and bypass the controller. You need wallets and policies that run remotely.
