Agent Platform Engineer (Control Plane)
Also listed as Platform Engineer, AI Agents
Working with MING as our Agent Platform Engineer, you build the control plane that turns a fleet of AI agents into something a CISO can sign off on and a client can rely on. This is the layer nobody has gotten right, not the model providers, not the big consultancies: skill and tool governance, cost control, audit trails, the security and isolation model, and the agent-building infrastructure that lets us stand up a new agent quickly and safely. It is also the layer we are betting the firm on. Model providers own the infrastructure. We own how agents are governed, made accountable, and made safe to put inside an organisation. That is your work.
You build the primitives every agent runs on: how skills and tools are registered, scoped, permissioned, and versioned; how we route models and control cost per agent and per task; how every agent action is logged into an audit trail that holds up to a client's risk and legal teams; how credentials stay out of an agent's reach so we can safely remove the human approval gate; and how a new agent goes from role brief to production without a hand-rolled setup each time. Looking ahead, we want to build our own agent harness to get deeper control of all of this, and you would lead that.
You report straight to Sebastian Mueller, Founding Partner, and work hand in hand with our Forward Deployed Engineers, who build client agents on the platform you own and surface the requirements that shape it. This is a build-and-own role with real autonomy. Sebastian scopes the problem with you and gets out of your way.
We offer you
- The chance to own the moat. Governance and audit are the part of agentic AI that is unsolved, and the part enterprises will not buy without. You would own it.
- Real autonomy. A spec-and-build role: you take a problem area and own it end to end, not a backlog of tickets.
- A clear forward frontier: leading the design of our own agent harness when the control plane is ready for it.
- A founder who builds the hardest parts beside you, and a founding role on a new function with a path to platform lead.
- High-leverage work. Every primitive you build makes every future deployment cheaper, safer, and faster.
- Flexible, location-open working.
Your specific responsibilities include, but are not limited to
Skill & tool governance
You make the agent's capabilities safe and managed.
- Build how skills and tools are registered, scoped, permissioned, versioned, and allow-listed across the fleet.
- Prevent capability sprawl and configuration drift. Make a new agent's surface area a deliberate decision, not an accident.
Cost control & model routing
You make the economics legible and bounded.
- Build model routing and the cost controls that hold spend per agent, per task, and per function.
- Give us cost-per-outcome as a real, reportable unit: the number we price and budget against.
Audit & accountability
You build the procurement-survivable layer.
- Build the audit trail that records every agent action and decision in a form a client can defend to risk and legal.
- Enforce decision-authority tiers, approvals, and promotions in the platform, not just in documents.
Agent-building infrastructure
You make standing up an agent fast and safe.
- Build the substrate that takes an agent from role brief to production: foundation, deploy gates, and the context and retrieval backbone agents draw on.
- Turn recurring patterns the field surfaces into durable, reusable platform capability.
Security & reliability
You make the platform safe and dependable by construction.
- Enforce least privilege and credential isolation. Hold tokens outside the agent so the human gate can come off safely.
- Own the reliability of the deployment substrate: authentication, secret rotation, deploy security gates, and the failure modes that take client agents down silently.
The harness (forward bet)
You lead the move to our own runtime when it is time.
- Pressure-test where an off-the-shelf runtime limits us and what an in-house harness would unlock.
- Design and build it when the control plane has earned the move.
Our ideal candidate brings
- Strong platform or backend engineering. You design systems other engineers build on, and you can own a domain end to end.
- A real governance and security instinct. Least privilege, credential isolation, audit, allow-listing, and a habit of red-teaming your own work. You think about what an agent should not be able to do.
- Comfort designing control-plane systems. Registries, policy, routing, cost accounting, audit, with the judgment to keep them simple.
- Evidence-first habits. You verify at the code and output level, and you treat "done" as a claim to check.
- The autonomy to ship unsupervised. You take a scoped problem and ship it without close supervision, the trait that actually recovers a founder's time.
- Fluency with context and retrieval substrates. Memory and hybrid search, the substrates agents depend on for accuracy.
Nice to have: Hands-on LLM evaluation and observability tooling (e.g. LiteLLM, Langfuse); experience with agent frameworks or runtimes (e.g. OpenClaw or similar) and an appetite to build one; container and fleet operations at scale.
Why MING
We are a small, senior team building the operating model for organisations where humans and agents share the org chart, and we run one on ourselves. The control plane you would build is the layer everyone needs and nobody has gotten right, and it is the layer our strategy is built on. If you want to own that, with real autonomy and a founder building alongside you, this is it.
Think this is you?
Send the AR department your work: a link to what you've built, and a few sentences on what you'd build with us. We read everything, and reply to the ones we'd want to work with.
Apply for this role
