Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
Meet The DevTool Founders Building For AI Agents
TL;DR
Agent experience is becoming a product requirement, not a nice-to-have — multiple founders said they now design CLIs, docs, exportable data, and “skills” specifically so Claude, Codex, Gemini, and other agents can use their products directly.
Founders are shipping far more code, even if they personally type less of it — one founder said coding agents now generate roughly 90% of the code, while teams are doing 10 features in parallel instead of one and reviewing rather than hand-writing most of the output.
The winning devtools may be chosen by agents before humans ever see them — Recall said coding models recommend its product surprisingly often, and several founders argued that if Claude or ChatGPT doesn’t surface your tool, it will become much harder to stand out.
What still matters is context, taste, and customer understanding — the sharpest line in the video is that “code doesn’t matter anymore”; what’s left is understanding problems deeply, having product judgment, and building real customer relationships you can’t “vibe code” around.
Documentation is being rewritten for context windows, not just human readers — one founder described compressing sprawling docs into agent-friendly formats that fit inside a coding agent’s context window, which led to higher integration accuracy and fewer mistakes.
The founders are bullish that AI raises the value of engineers rather than replacing them — examples ranged from engineering managers writing production code again to finance and support teams contributing logic via agents, while companies like Replit say they’re hiring more engineers, not fewer.
The Breakdown
A rapid-fire tour of the new AI devtools stack
The video opens with YC-style speed intros from founders building all over the agent stack: Infracloud on identity security, Resend as the “Stripe of email,” Recall for conversation data, Griptape reviewing more than 3 billion lines of code a month, Firecrawl for web data, Porter for compute, plus local LLM and subscription tooling. The common thread is obvious: these companies still serve developers, but more and more they also serve agents as first-class users.
Developers are still the buyers, but agents are becoming the interface
Several founders make the distinction cleanly: developers are the users, but agents are increasingly the distribution or even the customer. That shift changes product design — teams are building CLIs, skills, and machine-friendly interfaces so agents can operate their tools directly, not just through a human clicking around.
How AI changed the companies themselves
The conversation quickly moves from product surface area to org design. One founder says they started spinning up internal agents for marketing, sales, infrastructure, and DevOps; another says output has gone from one feature at a time to 10 in parallel. There’s also a strong emotional note here: if they were starting today, some would build around agentic infrastructure like sandboxes from day one, and others would get much better at ruthlessly deleting features that will be obsolete in a few months.
The old mistakes: drifting from developers and overbuilding polish
A few founders get unusually candid about where they got it wrong. One says every time they drifted from the “spirit of developers” — directness, speed, clear pricing — it was a mistake; another says they confused developer experience with perfect UI, when what actually matters is how fast users reach the aha moment. There’s also a nice product confession from a founder who says they leaned too hard on customer input and not enough on their own instincts, leaving the product with “crust.”
Docs, code, and the founder job are all being rewritten
One of the most concrete changes is documentation: instead of long, expository docs spread across pages, teams are rewriting them to fit inside an agent’s context window so integrations go more smoothly. On coding, the answers range from “I write prompts for humans and agents” to “every single day I push a PR,” but the pattern is clear: founders still code, just with Claude, Codex, and other agents doing most of the typing.
The surprising part: agents are good at strategy and even customer acquisition
The unexpected stories are the stickiest ones. Recall discovered that coding models recommend its product all the time, effectively turning agents into a new acquisition channel; another founder says agents were remarkably strong at critiquing roadmap plans after ingesting support tickets and recent PRs. Instead of replacing judgment, the best use here is pushback: founders bring the thesis, the agent spots what they missed.
Where this all goes next
The back half of the video gets more philosophical and more ambitious. Founders predict a world of many coding agents rather than one winner, with different interfaces for engineers, designers, support, and even finance; they also expect agents to handle incident response, procurement, access management, and much more. But the closing message is grounded: change is not slowing down, code itself is being commoditized, and the durable edge is context — knowing the problem, the customer, and what good looks like.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
