
Job Description
Lemma reads all of your agent conversations, finds the failures nobody knew to look for, and opens the PR that fixes them.
Why this role exists
For our first design partners we didn't sell software. We embedded. We sat in their Slack, read their traces at odd hours, learned their agent's failure modes better than they had, and opened PRs against their repo. Wordware shipped 8 PRs in one morning off issues we surfaced. Issues they didn't know they had.
It worked, and it doesn't scale as a founder activity.
This role is that motion, productized. You go into a new customer, get them to the moment where Lemma surfaces something that makes them sit up, and then take everything you learned doing it by hand and turn it into product so the next customer doesn't need you in the room.
If you want a clean separation between customer work and engineering work, this isn't it. The whole job is the seam between them, and roughly half of it is committing to our codebase rather than theirs.
What you'll do
- Own new customers end to end: integration, first traces, first surfaced failure, first merged PR
- Live in customer Slack channels. You'll know their agent architecture better than most of their engineers do
- Write the detection logic and evals that catch a given customer's specific failure modes, then generalize the ones that recur
- Open PRs against unfamiliar codebases and defend them in review to people who didn't ask for your help
- Be the only Lemma person in the room for the hard conversations. Security review, deployment architecture, "why should we trust your patches"
- Feed everything back into the product so the manual version of this job gets smaller every quarter
What we're looking for
- 3+ years shipping production software, plus something in your history that shows you're good in front of customers. Forward deployed, solutions engineering, consulting, founding, running your own thing
- Strong TypeScript. You'll be reading unfamiliar agent codebases constantly and writing patches for them
- You've built with agents and have real opinions about why they fail
- Genuinely fast. Our whole pitch is compressing the time between "something is broken" and "it's fixed." You should embody that
- Comfortable with the ambiguity of walking into a company you don't know and figuring out what matters in a week
- Bonus: you were the customer once. You ran agents in production and got burned
We move fast on hiring. Target is an offer within two weeks of first contact.
Onsite in San Francisco, with travel to customers. We sponsor visas.
Interview Process
We keep this short on purpose. Target is an offer within two weeks of first contact.
Intro call with a founder (30 minutes). What we're building, what you've built, whether the shape of the role actually fits what you want next.
Take-home project and deep dive. We give you a real problem from Lemma. You work on it on your own time, then we walk through it together. The conversation matters more to us than the artifact.
Paid work trial (2 days, onsite in-person). Real problem, real codebase, sitting with the team. You find out what working here actually feels like before you commit, which tells you more than anything we could say about it.
May ask for additional references.
Optimize Your Resume for This Job
Get a match score and see exactly which keywords you're missing
Job Details
- Category
- Software
- Employment Type
- Full Time
- Location
- San Francisco, CA
- Posted
- Compensation
- $150,000 - $200,000 per year
About Lemma
Lemma catches the silent, semantic failures your observability tools miss, where your AI agent looks like it worked but didn’t. We scan every trace to surface issues before users complain, identify root causes, and help you fix them without manual digging, so your agents improve over time.
More Roles at Lemma




Similar Software Roles



Found this role interesting?