TLDR Founders 2026-10-29

Ads get ignored on social media but not in TLDR! Reach founders, operators and startup leaders where they actually pay attention. Learn more about sponsorship opportunities.

Most companies don't need smarter AI models to unlock major productivity gains. The bigger bottleneck is context: models lack access to internal knowledge, unwritten processes, and the judgment sitting in employees' heads. Databricks is building a shared system of record that gives agents this context, alongside permissions and feedback on their work. This could reduce management layers devoted to moving information around, while shifting employees toward defining tasks, delegating execution, and judging results. The lasting competitive advantage would come from capturing what a company learns, so its AI systems improve even when it switches models.

Managing people used to be something you earned after years of grunt work. In this post, two people at OpenAI who were running teams of agents at 24 walk through where AI could end up on a company's org chart. In the simplest version, nobody changes seats, so everyone just becomes a manager (an accountant still owns the accounting while agents do the spreadsheet work). The accounting team doesn't shrink either, because the rest of the company is getting more productive and needs more accounting than before.

Most customer dinners are planned around the night itself, and the "great meeting you" email afterward gets ignored. One conference dinner turned into $4M in revenue for Recall.ai. The trick was to put a rep at every table whose only job was seating customers next to prospects, so prospects heard about the product from people who already trusted it. Reps then sent each prospect an intro to the customer they sat beside ("You two were talking about X at the dinner"). That email is much harder to ignore than a thank-you note.

Companies spend about $6 on services for every $1 on software, so a lot of AI startups now want to be the law firm instead of selling it software. Selling the work itself might only be step one. Big firms are building their own AI (Kirkland & Ellis is putting $500M into its stack), so give the service away and make money on the deals and data that run through it. For example, Rockwell runs a free benefits concierge for employees at mid-market companies, so it sits right in front of their healthcare spending.

Claude Smart reports surface what a team gets done with Claude and what that work costs. Admins can now create surveys to get feedback on how the team is working with AI. Anthropic has introduced Pool group budgets, giving teams a shared monthly budget. Member limits are still available as a guardrail against unusually high individual use.

Vercel now shows how AI bots use your site, including which pages assistants fetch when someone asks them a question and which pages return errors to those bots. Querying the numbers from the CLI requires Observability Plus.

The Slack Code API allows agents to start code channels, add relevant people and files, post updates as they work, and archive channels when the work is done. It enables teams to work together with agents on tasks with the same shared context. Users can see exactly what agents are doing and who approved it, and workspace controls and data boundaries are built in. Users can choose what agents and models they want to use to suit their needs.

AI companies need a new way to pitch their economics. The cost of serving customers is much higher with AI. Companies need to separate gross margin by customer and gross margin by pricing cohort. When changing pricing, you ideally want to show investors that customers on the new model have higher margins than those on the old one, and that the gap keeps widening.

Spawn is a company with 10 people that spends 30 times its payroll on AI tokens. The company ships almost as much as Anthropic, with 77,000 substantial changes in the last 30 days. Its work is done by a fleet of permanent AI agents that each own a part of the company. Each new model improvement has made the team faster and more coherent, and pointed the company's velocity more in the right direction.

Being named is easy, but being described correctly requires a bit of effort.

One acquisition target was turning away work worth 4–5 times last year's revenue because it couldn't afford more staff.

Review what work a proposed moat would require every week, and then decide whether customers value the result enough to support that work.

Wispr Flow pays affiliates $35 to bring over paying Granola users, while Granola pays $20 to recruit new ones.

System of Action is a series of essays that details what it takes for vertical software to win in the age of AI.