Here's a question most developer tool companies can't answer: which AI coding assistant writes the best code for your platform?
We couldn't either — at least not in any rigorous way. This is increasingly an existential problem.
Two years ago, you would've been fired for shipping AI-generated code to production. In the not-too-distant future, you might be fired if you don't.[1]
The stats on adoption of agentic coding tools underscore the magnitude of the problem: 85% of developers now use AI tools regularly, and 62% rely on a coding assistant.[2]
So, if the primary way developers experience Temporal is increasingly through an AI agent, shouldn't we be optimizing for that?
The persona is an agent#
Traditional developer experience optimizes for humans: people reading docs, following tutorials, copying sample code. But that's not how it works anymore. Developers prompt Claude Code or Cursor or Codex, then the agent consumes context on their behalf and suggests a course of action.
This changes every part of the development lifecycle: learning the platform, integrating the codebase, troubleshooting, and operating in production.
Which means the experience of the agent is inextricably linked to developer velocity.
What we're doing about it#
We're hiring a PM whose entire job is making AI coding assistants better at Temporal. This is a dedicated role with dedicated metrics. And it's different from any other PM role on the market.
The work looks like this:
- Building a benchmark suite that measures how well various LLM coding assistants perform on real Temporal tasks, both coding and operating expertise — think SWE-Bench, but for our platform.
- Creating and validating assets (agents.md, Cursor rules, Claude skills, knowledge packs, docs, MCP servers, etc.) that demonstrably improve output quality.
- Closing the feedback loop between what developers struggle with and what we ship to fix it.
The key word is demonstrably. You can't iterate efficiently on vibes.
The cadence will feel familiar to anyone who's done growth PM work: hypothesize why a particular tool is failing in a specific way, build an intervention, measure whether the benchmark moved, and iterate. Fast cycles. The difference is the "user" you're optimizing for isn't a human clicking through a funnel — it's an LLM consuming context and generating code.
A different kind of PM#
This role inverts the traditional PM skillset hierarchy. Here's how we're prioritizing:
- AI coding tool proficiency (most important)
- Rigorous analytical thinking
- Product management fundamentals
- Temporal knowledge (certainly nice to have on day 1, but you can learn this)
Put simply: we'd rather hire someone with excellent instincts for getting the best out of AI tools who can learn our specific domain, than a seasoned PM who's "interested in AI."
Why this matters beyond Temporal#
I expect most developer-focused companies will have a role like this by the end of the year.
The opportunity is straightforward: every improvement compounds across every developer using AI tools. Fix one shared context file, and every developer using that tool benefits. The leverage is enormous. If you're not intentionally optimizing for this, you're not addressing the full developer experience.
We're looking for someone who lives in Claude Code, Cursor, or Codex and is able to finesse better performance out of them. If that's you, we'd love to talk.
Footnotes:
[1] To be clear, I'm not suggesting developers will disappear — I work at Temporal because I think the profession will only get more valuable. Nor am I saying that we'll just start blindly committing whatever comes out of the model. But, willfully resisting these tools will hurt your career; at least that's what every engineering leader I've talked to has said.