Garry Tan
73.4K posts
@garrytan
President & CEO @ycombinator —Founder @garryslist—Creator of GStack & GBrain—designer/engineer who helps founders—SF Dem accelerating the boom loop
"I realized tech is this thing that can bring people out of whatever situation they're in and often into prosperity. And that's what I want for everyone." @ycombinator’s @garrytan tells @emilychangtv how tech changed his family's life. Watch here: trib.al/sxg1VGR
Today is May 25th, 2026. This is the first time I remember reading an LLM-produced public artifact which is obviously professionally relevant and which is sufficiently complete that I do not perceive the lack of a human author materially compromising its utility to me.
1/5 I'm a cardiologist. I have spent twenty years watching cholesterol destroy arteries, trigger heart attacks, and kill people I care about. Today, Eli Lilly presented data that may begin to end that era. VERVE-102. A single infusion. One dose. It uses base editing to permanently turn off the PCSK9 gene in your liver. Presented today at the European Atherosclerosis Society Congress: 88% reduction in PCSK9. 62% reduction in LDL cholesterol. Sustained up to 18 months. No treatment-related serious adverse events. One infusion. Not daily pills you forget to take. Not monthly injections. One dose — and your cholesterol may stay low for the rest of your life.
Right now I just use my personal AI and our company brain and it screws up and I tell it to fix it and write tests for it. Also I do cross modal evals on progressive batches (eg if there are 10000 items do 5 and eval the input and output and skill, then keep doubling the batch size as you go)
Gradient descent for SKILL.md files sounds interesting, maybe a bit complex but it's becoming a real part of agent harness. SkillOpt is one of the first papers to treat markdown skill files as trainable parameters and provides a proper optimization framework for them. A few things I learned that you should consider too. 1. The validation gate is the only thing that matters in a self-editing loop. Held-out set, strict improvement, ties rejected. End-to-end, their best skills land with 1 to 4 accepted edits total. If your "self-improving agent" is accepting most of what it proposes, you're shipping slop. 2. Bounded edits are better than full rewrites. 4 to 8 edits per step is the sweet spot. Remove the budget and performance collapses. This is the textual analog of learning rate, and it transfers to any LLM-as-author loop. If you're using an agent to refactor your docs, your prompts, or your skills, cap the diff size. 3. Compactness wins. Median final skill: ~920 tokens. Skills do not need to be long. They need to be high-signal. Most skill files I see are bloated because length feels like effort. It isn't. 4. The harness is becoming less important; the skill is becoming more important. A Codex-trained skill ported into Claude Code hit +59.7 points on SpreadsheetBench. Procedural knowledge is more general than the runtime that produced it. 5. Frozen model + trained context is the practical adaptation. GPT-5.4-nano with a SkillOpt'd skill ≈ frontier behavior on procedural benchmarks. Cheaper, portable, inspectable, zero inference-time cost. This is the answer to "how do we adapt a frontier model for our domain" for almost everyone who isn't training their own models. 6. Verification is the bottleneck. Every gate in this paper depends on an auto-grader. That works for benchmarks. It fails for writing, design, and strategy, exactly the open-ended work we want to automate. Whoever builds the verifier for open-ended tasks owns the next stage. There are also two leassons I learned while shipping v2.3.0 of my Context Engineering Agent Skills repo, measured across composer-2, claude-opus-4-7, gpt-5.5, and gemini-3.1-pro via the @cursor_ai SDK: - Description and body are two different surfaces. The router only sees the description. The agent sees the body once activated. They can quietly disagree, and only end-to-end task tests catch it. - Aggregate accuracy is the wrong unit. When I rewrote three descriptions, the corpus average moved ~1pp. Individual skills moved 23–25pp. Per-skill effect size is where the action is. Also, in Feb 2026 I shared a piece called Personal Brain OS arguing that the markdown file is a first-class substrate for agent state. SkillOpt is the optimizer-shaped version of that same argument: not "store memory in files" but "treat files as trainable parameters with proper optimization machinery around them." That's the move from static to measured. The fast/slow split they describe already lives implicitly in the digital-brain-skill repo: - voice-guide and tone-of-voice.md are slow-state (rarely touched) - posts.jsonl and bookmarks.jsonl are fast-state What SkillOpt adds that I didn't have is a protected section invariant, a structural guarantee that fast edits cannot overwrite slow lessons. Removing that mechanism cost them 22 points on SpreadsheetBench. Worth borrowing. If you're building agents, SkillOpt: Executive Strategy for Self-Evolving Agent Skills is a good paper to read: arxiv.org/pdf/2605.23904
New research from Microsoft Research I see a lot of AI engineers handwriting agent skill docs and hope they generalize. Probably not optimal. This works show why. It treats the skill doc as a trainable external state of a frozen agent instead. It introduces SkillOpt, where an optimizer model makes validation-gated edits to the skill file. It adds, deletes, or replaces instructions, with a textual learning rate that controls how aggressively each round rewrites the doc. The agent itself never changes. SkillOpt is best or tied on all 52 (model, benchmark, harness) cells. On GPT-5.5 it adds 23.5 points in direct chat, 24.8 with Codex, and 19.1 with Claude Code over no skill. It beats human-written skills, TextGrad, GEPA, and EvoSkill, carries zero extra inference-time cost, and the learned skills transfer across models and harnesses. Paper: arxiv.org/abs/2605.23904 Learn to build effective AI agents in our academy: academy.dair.ai
YC deadline has been extended until this weekend. If you are a founder who missed it before, now is the time... Also, you now get $2M worth of OpenAI tokens if you're selected! (screenshot of @agupta's tweet, gp at @ycombinator)
CEOs are the most delusional about AI. Detached from reality.
Possibly the thing we will most realize looking back: intelligence was so big that lots of companies were going to succeed. It's not so simply bucketed into chatgpt and claude code.
These concepts coming soon to GBrain this week
