Build the workbench for delegated work.
PACKWOLF is a closed-beta AI agent workbench for teams that need recurring work to run with memory, approvals, tools, and an audit trail. We hire slowly because every seat changes the product.
Closed beta
Small team, narrow surface area, direct customer feedback. The work is still close enough to the metal that every person changes the product.
Builder-led
Engineering, design, and GTM sit near the customer problem. The best work here turns support threads and traces into product changes.
Remote-aware
We care about written context, clean handoffs, and people who can move work without needing a meeting to remember why it matters.
The culture is in the operating details.
The careers page should tell you what the work will actually feel like. This is the current contract.
Read the load-bearing code.
The critical path has to be inspectable. If you own a subsystem, you can explain the failure modes, the logs, the escape hatches, and the parts we still do not trust.
Write the thing down.
Specs, incident notes, customer learnings, and product decisions should survive the week. Documentation is not theater here; it is how a small team keeps memory.
Match approval to stakes.
We ship tools that let agents act, so our own work uses the same habit: quick on reversible changes, deliberate when data, money, credentials, or customer trust are involved.
Make the trace better.
When something breaks, the product should help the next person understand it faster. We prize engineers and designers who turn confusion into surfaces, logs, tests, and docs.
Good fit is specific.
We are building infrastructure for AI work that touches customer data, external tools, schedules, and approvals. The best people here are comfortable with both product taste and operational consequence.
You like ambiguous systems.
Agents, memory, workflow execution, model routing, permissions, and UX all touch. You can hold the whole shape without pretending it is simple.
You care about operators.
The customer is not buying magic. They are trying to trust delegated work. You notice where a human needs control, proof, or a clean way out.
You can make small teams faster.
You reduce coordination cost. You leave clear artifacts. You spot when a process is protecting quality and when it is just slowing motion.
- You want a mature ladder, a large peer group, or a role where the edges are already sanded down.
- You prefer building only the shiny surface and leaving observability, migrations, docs, or support paths to someone else.
- You need every answer to be known before you start. We make careful bets, but this is still early product work.
- Competitive cash and equity for the stage.
- Remote-aware work with written context and direct access to decision makers.
- Tools, hardware, and model credits needed to do the job well.
- Flexible time off, with the expectation that adults protect their energy before the work gets brittle.
- Benefits details are finalized per role and geography as we leave closed beta.
Profiles we always want to meet.
We are not running a broad recruiting pipeline today. These are the people we still want to hear from when the evidence is strong.
Agent infrastructure engineer
You have shipped memory, observability, scheduling, sandboxing, eval, or tool-execution code for AI runtimes and watched it survive production. Bonus if you can show a trace, postmortem, or hard-won debugging story.
Start a conversationProduct designer for dense systems
You can design an activity trace, approval queue, audit log, or workflow editor without turning it into a dashboard soup. You care about information hierarchy, product taste, and the difference between power and clutter.
Start a conversationOperator-first GTM builder
You have sold or supported infrastructure, devtools, observability, AI tooling, or security products without dumbing the product down. You can run a beta cohort, write a crisp customer update, and bring product signal back intact.
Start a conversation
No mystery funnel.
We keep the process closer to working together than performing interest. If the role is not real yet, we will say that.
- 01
Signal
Send the strongest evidence you have: shipped work, writing, traces, demos, customer outcomes, or a crisp note on why PACKWOLF is your kind of problem.
- 02
Working conversation
We talk through something real from the product: a trace, a workflow, a security boundary, a design surface, or a customer problem.
- 03
Paid work sample when useful
For roles where a sample helps both sides, we keep it bounded, compensate it, and use work that resembles the job.
- 04
Decision
We tell you where we are, what we learned, and whether there is a real seat now. We do not keep people in a performative pipeline.