Skin in the game

Right now the name of the game is trial and error. There are no hard and fast rules for how to use AI that apply to every organization. The "rules" we think we know are often fleeting; there is only the wisdom born from experimentation and experience.

Which is great news. No matter how far behind you feel (and seemingly everyone feels behind right now) the main thing standing in your way is a willingness to try things, to fail some, and to learn. And remember, you're not going to get organizational transformation through individual risk and individual learnings, so we're talking about org-level experimentation.

Don't dismiss the learnings of those around you because they come from a smaller company, are in a different industry, or seem too extreme. Be curious. Maybe what they did doesn't apply cleanly to your org, maybe it's not right for you (but maybe it is?). How can you know unless you're willing to put real skin in the game?

Here's what we've been reading, writing, and saying this week.

Making Our Monorepo Ergonomic for Agents

From the Atlas team at Basis. Detailed piece with lots of tactical tips that drives home the importance of making sure your codebase is agent-friendly. We love that they have actual headcount dedicated to improving organizational productivity with AI (we think this is a must!) and we love it even more that they're sharing their learnings out loud.

Send it to your engineering AI champion and see if it sparks anything — there's bound to be at least one improvement you can apply to your codebase in there.

Notes from a Real CHRO

Mike Hudak posted notes from a call with the Chief HR Officer of a $500M consumer business. This echoes what we're seeing and hearing as well: most companies only have a "tool rollout" playbook with no formal AI strategy; they're seeing extremely uneven adoption (a few individuals way ahead and everyone else trailing way behind, what Hudak calls the "haves and have-nots") and a lack of org-level gains.

If you read the Basis piece and instantly felt behind, this is the real baseline. Most orgs are here. You're not alone, but things are changing rapidly. The orgs that make space for experimentation and learning now will gain a compounding advantage.

The 100x Org

ClickUp's CEO on their decision to lay off 22% of the company on Wednesday. And whether you agree with the layoff math or not, it's a banger of a read.

I only see two options: wait for this to play out gradually in the market or be honest about what I'm seeing and act proactively.

The most interesting part? The idea that they're going to restructure the company around higher-leverage ICs and introduce salary bands that go up to $1M cash. If the CHRO post describes a shared feeling of being lost in the wilderness, ClickUp joins the list of companies being proactive about charting a path out.

Three companies, three different answers. Yours probably won't look like any of them — and probably can't. The work is in trying anyway.

Pick a model. Don't marry one.

Claire on X this week. Anthropic has done an unreal job at papering the earth with enterprise contracts. Every company we walk into "just went all in on Claude," while every cutting-edge builder we know has already moved to Codex. File this one next to the Interoperability principle from Basis. Same idea: don't bet your stack, or your org, on a single vendor right now. To the fast and flexible go the spoils.

Three things to try this week

• We already gave you one — read the first post above from Basis and make one improvement so your codebase provides better standalone context for agents.
• [Lapdog](https://lapdog.datadoghq.com/) — new free tool from Datadog that traces what your agent is actually doing inside Codex, Claude Code, and Pi. Runs locally, so you can try it without convincing procurement first.
• [HTML is the new markdown](https://www.chatprd.ai/how-i-ai/claude-code-anthropic-thariq-shihipar-on-replacing-markdown-with-html) — Thariq Shihipar from Anthropic on "How I AI" talking about Claire's new favorite life-changing trick: have your agent generate HTML instead of markdown for artifacts you'll actually read and comprehend.

Happy reading, and for those in the U.S. have a great long weekend!

— Claire + Zach