Skip to content

Build the runtime
agents run on.

Constal is the runtime for production AI agents: durable sessions, sandboxes and browsers, governed tools, Policies and approvals on every effect, and evals that show whether an agent is getting better.

LOCATION
San Francisco Bay Area, in person
TEAM
Small team, whole-system ownership
SHIPS
Every day, gated by evals
CONTACT
careers@constal.ai

Open positions

1 OPEN ROLE
Open positions at Constal
TITLETYPELOCATIONDETAILSAction

Technical Founding Engineer

Durable execution, isolation, authorization and evals

 FULL-TIME SF Bay AreaIn person $160K–$250K+ 0.5%–1.5% equity VIEW ROLE

Not the right role? Write anyway and tell us what you would build on Constal.

How we work

01

A small team with a large surface

Each engineer owns whole systems, from the API and SDK down to Durable Objects and Drivers.

02

We ship every day

Staging first, then production. Evals, not meetings, decide when a change is ready.

03

Agents write much of our code

Held to the same bar as anything else: tests, evals and review.

04

In person

Our hard problems are design problems, and those move faster at a whiteboard.

Models keep getting better at reasoning. What keeps agents from doing real work is everything around the model: state that outlives a process, Credentials an agent can use without ever seeing them, actions a person has to approve, spend that needs a ceiling, and a record of what happened that you can replay. That is the layer we build.

We run our own agents on Constal in production, including a coding agent that takes work from Slack and GitHub, so we hit the platform's sharp edges before our customers do.

Technical Founding Engineer

APPLY BY EMAIL

Location: San Francisco Bay Area, in person
Type: Full-time
Compensation: $160K–$250K + 0.5%–1.5% equity

You will be one of the first engineers at Constal, working directly with the founders on the core of the platform. This is a systems role more than a prompting role: the work is durable execution, isolation, authorization and measurement for software that makes its own decisions.

What you will do

  • Build the execution core: the sessions, Runs and journal that let an agent stop, resume and replay exactly, at scale.
  • Build the Resources agents act through (models, sandboxes across providers, browsers, memory and skills) and the Drivers behind them.
  • Make it safe to give agents real authority, with Policies, approvals, scoped Credentials, egress controls and budgets.
  • Build the eval and training loop: datasets drawn from production, scorers and judges, and the path from a failed Run to a better agent.
  • Work with the developers building on Constal, and turn what they run into platform primitives rather than one-off fixes.
  • Help hire the next engineers and shape how we build.

You might be a fit if

  • You have built and operated distributed systems in production, and you can tell us about the failure that taught you the most.
  • You have built agent systems that go beyond calling a chat API: tool execution, state across turns, retries, evals and replay.
  • You write TypeScript fluently and are comfortable going down a layer (runtimes, storage, networking, containers) when the problem lives there.
  • You use coding agents every day and have opinions about where they fail.
  • You want ownership and ambiguity, and you would rather design the missing primitive than patch a symptom.

Nice to have

  • Experience with Cloudflare Workers and Durable Objects, or another actor or serverless runtime.
  • Sandboxing and isolation: containers, microVMs, network egress control.
  • Authorization systems, policy engines or secrets management.
  • Evals, reinforcement learning or fine-tuning pipelines for language models.
  • Time as an early engineer or a founder.

We care much more about what you have built than where you have worked. If this sounds like you but you do not match every line, apply anyway.

Interview process

  • Intro call. Thirty minutes with a founder about what you have built and what you want to build next.
  • Technical deep dive. Walk us through a system you designed and the decisions behind it. No algorithm puzzles.
  • Paid work trial. Time with us in person, on a real problem in our codebase.

How to apply

Email careers@constal.ai with a link to something you built and a few sentences on the hardest problem you solved in it. A repository, a write-up or a demo all work. A résumé is optional.

APPLY BY EMAIL

Public notices

Labor Condition Application (LCA) notice (PDF)