Building product @get_writer. Previously @ateamsinc, @brexhq, @flexport, @square, @gumroad, @stripe. Member @sandboxers and @globalshapers
San Francisco/Jerusalem
Joined March 2013
- Tweets4K
- Following857
- Followers593
- Likes41.8K
Enterprise workflows aren't waiting on smarter agents. They're waiting on someone with the authority to delete step four of the approval chain.
This is correct. The vast majority of enterprise workflows have not actually had agents applied to them in any meaningful way. The economy probably has this deployed < 1% right now.
At best you have a few automated systems in a company that do a single pass of an LLM to make a decision or provide feedback, but very little real “automation”. Some of this is a statement about the adoption rates of technology in enterprises, but the vast majority of it is really just due to the capability of AI itself.
That’s all changing. Now with computer use, long running agents, harness improvements and surrounding infrastructure, we’re finally at a point where we can start to apply agents at more complex workflows.
We will see self-improving systems that can take any particular workflow goal and improve the execution rates over a volume of instances. And with the emerging form factor of agent swarms, you can just choose to deploy any amount of compute at a problem and agents will off and do the work.
This won’t be about and users prompting agents to complete a specific task for them, but instead the organization reengineering the underlying workflows to take advantage of agents in their processes. This will have a massive impact in finance, life sciences, healthcare, and other information-heavy domains. Lots of opportunity to build and deploy these systems right now.
Every company rents the same model. The moat is your company remembering why it said no last time.
Every company can rent the same frontier models, so the edge comes from what a company's agents learn about its own business. Databricks' answer is a system of record built for agents.
Before an agent can be trusted with real work, that record has to tell it:
1. What's true: the company's own data
2. What it means: how the business defines things (grounding agents in it raised accuracy from about 50% to 84.5% in Databricks' testing)
3. What it's allowed to do: an identity, a budget, and guardrails
4. Whether it worked: grading against the company's own definition of good
Feed every correction back into the record, and the company keeps what it learns even when the model changes.
"If you replace the AI model at your company and the new one inherits your definitions, permissions, evaluated cases, customer history, and hard-won exceptions, that is a foundation for a sustainable competitive advantage."
From the latest Profiles of the Future by @bhorowitz and @mikemcg0 on Databricks & the Future of the Firm: a16z.news/p/databricks-and-t…
I wrote about this as the living brain of the enterprise: matanpaul.substack.com/the-l…
Labor is priced on scarcity. A great engineer costs what they cost because there aren't enough of them. When every company can rent the same intelligence, nothing is scarce, and pricing it like labor falls apart.
Dario Amodei: some tokens Claude produces are worth a few cents, others could be worth tens of millions of dollars.
His examples:
1. Telling someone to restart their Mac: the model has said it 10 million times, worth a dollar or less
2. Telling a pharma company to move an aromatic ring to the other end of a molecule: potentially tens of millions
3. Expect pay-for-results pricing, or labor-like pricing by the hour
4. It's a new industry, so many models will be tried
Today pricing is simpler: per the S-1, usage-based pricing was 80% of H1 2026 revenue and subscriptions 20%.
Full breakdown here:
theaiopportunities.com/p/ant…
Amjad's framing is the right one: humans exercise judgment, agents execute. The hard part isn't the habitat. It's mapping who decides what. That map is the firm now, and almost nobody has drawn it. I wrote about it as the orchestration graph.
matanpaul.substack.com/the-o…
For centuries, the architecture of the firm has barely changed: hierarchies of people organized into departments and coordinated through bureaucracy.
AI agents are about to upend that.
We’re moving toward the self-driving company where humans exercise judgment and agents execute most of the work.
At Replit, we’re building the foundation for this new kind of company. We’re calling it “the Agent Habitat.”
Marc Andreessen wrote a fantastic piece placing this shift in the history of computing. Every major computing revolution began with chaos before businesses figured out how to harness it.
Ethan's right, and it's worse than it sounds. We kept humans in the loop by letting them read the model's thinking. Take the trace away and the human is approving outcomes they can't see. That's not supervision. It's rubber stamping
My AI chief of staff needs to consult my AI general counsel. I just asked it to book you a haircut.
למי שלא מכיר זה אותו פלייבוק he.wikipedia.org/wiki/%D7%94…
הם מפחדים!
Ten AI subscriptions, twelve computers, and a proxy to keep them signed in. At some point you stopped having an assistant and started running its IT department.
The more of my life Google holds, the less I want to experiment with its AI. “One account for everything” gets less appealing when a chatbot can trigger a policy warning.
Just got several alarming, scary, and ridiculous notifications from Google. Apparently, I used Gemini in some way that violates Google policies, and if I do it again, I may lose access to unspecified services and content.
At the time of the notifications, I was driving and not using Gemini at all, to my knowledge.
What did I do? I have no idea, the email doesn't specify any details. This is scary enough for me to stop using Gemini permanently. Risking my whole account because some AI thinks some query is not kosher for whatever imaginary reason is not worth it.
"Your account has been used in a way that violates Google's policies
It looks like a Google service has been used in a way that violates Google's policies. Here are more details:
Service: Gemini
Issue: Trying to interfere with or disrupt Google’s services or infrastructure, or attempting to use Google services for disrupting others’ infrastructure or services
Your access to Google services hasn't been restricted at this time, but if you continue to violate our policies, your account may lose access to specific services and content from those services. If that happens, you'll get an email informing you about the restriction and the option to appeal any violation you've been informed about. Visit our help center to learn more.
Please review our policies to make sure you use Google services safely in the future."
Retention isn’t loyalty. Sometimes your customers just didn’t have an easy way out.
AI is going to expose a lot of “great retention” as expensive captivity. Before you raise prices, know whether you’ve earned their business or just made leaving painful.
When customers can build the part of your product they actually need, you’d better know what your SKU is worth to them.
Ok my dot is done. I have a @Plaid account (free) and dot can get my Amex checking and CC charges every week, map them to IRS categories using the Microsoft Decision-1 model, store them and ask me about anything new. Since all transactions fit to a category it's pretty simple. Then at the end of the year I give my accountant a PDF and CSV.
It's also fully automated.
$240 saved.
We automated Christmas and still gave all the work to Mrs. Claus.
Christmas is only 61 days away & this year I'm using AI to keep all the gift stuff under control.
@clairevo made tradBot and @SoniaBaschez talked about needing an AI wife so for this holiday season I present: Mrs. Claus. 🎄🤶🏻🎁
She's a @bot template that does the following:
-Gets onboarded by asking who you are shopping for, budget, and what inboxes to connect to for receipt scanning
-Builds your gift Page, which is a living doc in your Library that is updated with item, amount, quantity, etc per person
-Every day she scans your inboxes and asked if stuff you've purchased should go into the Page
-She also scans your inbox daily leading up to Cyber Monday for all the newsletters your are subscribed to to find deals relevant to your gift recipients
Give her a whirl & lmk what you think! x.ai/bot/FTZD8KMeI78-iYxiDla…
An AI that treats “no appointments available” as the start of negotiations. Finally, proper localization for Israel
Thousands of years of Shabbat, but Israelis still needed Americans to tell them it was cool.
I'm sure the food at Neta's was better then what we have in SF
ארוחת טק-שבת ת״א 1: היה מדהים! ☺️
פחות מחודש אחרי שעלה לי הרעיון לארגן ארוחות שבת ליזמים בת״א, אתמול היתה הארוחה הראשונה והיה פשוט מדהים! להיות בחדר אחד עם אנשים מעניינים שבונים, משקיעים וחושבים על הטכנולוגיה של המחר עם המון אהבה ותשוקה בעניינים - בדיוק מה שרציתי!! :)
תודה מיוחדת לשחר כהן (@S_y_Cohen ) ול-Aurelius Capital על התמיכה והספונסרשיפ של המפגש ולתהילה וזיו על האירוח המקסים.
זאת רק ההתחלה - בקרוב פרטים על הארוחה הבאה! 🍷🥖
@orslimy @advahandelman @TPelled @Yael_daihes @shayahal1 @KachlonKfir @YochaiPagi @BambaDror
@nir_altman @z_bakhaji @orenyomtov
Apparently Slack didn’t capture enough of your interior life. Here comes the AI chat logs
things that eng leaders are wondering about rn re: ai
- fml cyber risk
- should we force convergence on tools
- will engineers riot if i read their chat sessions
- how do i forecast token spend for 2027
- verification, verification, verification
- what should non techincal ppl build agents on
- vm + local execution + connector auth, oh my
- bug triage
- pr review
- what is even a spec
- making design systems more programatic
- fml cyber risk
what am i missing?
→ @cxodev
The model says "two to three days" because we trained it on ten years of Jira tickets. It didn't learn how long the work takes, it learned how long engineers say it takes.
Opus 5.5 is the best work estimator model i've ever used. This is calibration that is increasingly important as we delegate more and more work — don't tell me that it's going to take 2-3 working days (what a human would do) when for you, an enthusiastic, indefatigable, and embarrasingly parallelizable AGI, can get it done in 2-3 hours.
