TLDR Founders 2026-05-29

If your AI bill is growing, the real question is no longer “are people using it?” It is “what business outcome did those tokens buy?” The useful idea here is token-to-outcome tracking: cost per resolved ticket, processed claim, reviewed contract, avoided hire, or dollar of revenue moved. For founders selling AI into companies, usage is not enough. The buyer will soon want proof that inference spend turns into work that matters.

AI is transforming software by replacing static workflows with adaptable intelligence, creating a need for robust AI harness systems. These systems comprise seven components: context retrieval, tool integration, orchestrated workflows, state management, secure computing sandboxes, observability, and cost optimization. Companies effectively managing these elements will dominate the emerging competitive landscape in software.

Gokul Rajaram contrasts his Facebook days where engineers shipped multiple daily updates without explicit QA gates versus Square where the team pushed one app update every few weeks behind human QA, alpha, and beta gates, arguing the difference was a deliberate cultural trade-off rooted in how critical each product was to customers' lives.

The requirement to understand the user was true when SEO was purely about traditional search. It is even more true now that AI has substantially altered the user journey. Businesses need to confirm whether users who would benefit from what they have to offer are actually searching for it, and whether what they find when they arrive matches the performance the business delivers. If either is not true, they need to redirect their search budget.

A good demo can be the difference between success and failure. This post presents a list of what sets S-tier demos apart from the rest. It covers the basic structure of a demo, storytelling tactics, setup and delivery, and more.

Plane co-founder Staszek Kolarzowski (YC '17) will give live demos of the AI engine he built to run sales targeting, call coaching, and commissions that scaled to $Ms in ARR in this live webinar. Join to see what's in production and the hard-earned lessons that shaped his systems. Replay available. Save your spot

Anthropic has introduced a new dynamic workflows feature in Claude Code that helps Claude take on challenging tasks end-to-end. Dynamic workflows can handle problems too big for one pass by a single agent, dynamically writing orchestration scripts that run tens to hundreds of parallel subagents. It is available now in research preview for users on Max, Team, and Enterprise plans and on the Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry. The feature consumes substantially more tokens than usual, so users should start on a scoped task to get a feel for usage.

Amazon OpenSearch Serverless is a fully managed search and vector engine designed to build agents. It now auto-scales 20 times faster and provisions resources in seconds. The latest update introduces complete decoupling of compute and storage through a new shared storage layer, so customers can scale compute up and down independently. OpenSearch Serverless now features native integrations with AI development platforms, so developers can provision search infrastructure directly from their development environments using natural language commands.

The defensibility question can become a trap at seed. Most great companies do not start with a perfect moat. They build one by learning faster, hiring better, earning trust, and executing for longer than everyone else. The founder takeaway is not to ignore moats. It is to know when “what stops OpenAI from building this?” is a real risk, and when it is just a lazy way to underrate speed, taste, and customer obsession.

The General Partner model and the startup model are not the same game. Startups are supposed to chase huge, ambitious targets, knowing that most who try will fail. General Partners run diversified portfolios specifically so they don't lose their liquidity providers' money. Many founders grind themselves down trying to get their investors some kind of return when the investors have already written the company off a long time ago. These founders are better off shutting it down and figuring out the next thing.

A full-time hire is a bet that the same work will keep showing up.

YC CEO Garry Tan reframes what Y Combinator actually delivers by borrowing psychologist Alice Miller's concept of the "enlightened witness," the single strongest predictor of whether traumatized children grow into functional adults: not someone who could fix or rescue them, but someone credible enough to confirm their reality.

Unusual Ventures Director of Programs & Community Niamh O'Donnell shares the seven-step playbook from arriving in San Francisco at 24 on a one-year Irish visa with no tech network to hosting rooms full of founders, investors, and Fortune 500 leaders five years later.