Baseten's engineers want to work in an AI-first way. What's missing isn't enthusiasm — it's the platform underneath it. Today everyone assembles their own agent config, context files, and MCP servers, so the good patterns stay trapped in individual setups instead of becoming defaults everyone inherits
You'll build that platform: the agent configurations tuned to our monorepo, the context and tooling layer that makes agents competent in our codebase, the evals that tell us which approaches actually work, and the rollout mechanics that get a new engineer productive with agents in week one
You are not here to mandate how engineers use AI — you're here to make the good path the easy path. Success looks like teams adopting what you build because it beats what they'd cobble together themselves, not because a policy requires it. Platform engineer, not AI evangelist. Ship infrastructure, measure it, kill what doesn't work, let adoption be the referee
The playbook for AI-first SDLC doesn't exist at any company yet. You'll write ours
WHAT YOU'LL BUILD
Agent substrate — Repo-level context infrastructure that makes agents competent in our codebase (CLAUDE.md/AGENTS.md conventions, architecture and domain context, and the tooling to keep it accurate as code moves). Internal MCP servers giving agents scoped access to CI, observability, incident tooling, deployment state, and docs. Shared skills, subagents, and hooks that encode Baseten workflows. Sandboxed environments where agents can build and test safely
The golden path — Project templates and onboarding that ship with AI tooling configured and working. Self-serve infrastructure so teams build their own agents without you as the bottleneck. Gateway, auth, cost controls, and audit logging for internal model access
The feedback loop — Eval harnesses that answer "is this config better than that one" against real Baseten tasks, not vibes. Instrumentation of AI tool usage and its downstream effects on cycle time, review latency, and change failure rate. Honest reporting, including on what you built that didn't pan out
Agents in the SDLC — Automation where agents earn their keep: PR review triage, test gap-filling, incident context assembly, migrations and refactors, codebase Q&A. Integrating agents into CI/CD with guardrails that make it trustworthy
Own the internal AI developer platform end to end — architecture, build, rollout, operation, measurement
Evaluate and integrate third-party AI coding tools (Claude Code, Cursor, Codex, and whatever ships next quarter), and build the context layer that makes them work against our monorepo
Build frameworks that let other engineers create their own agents without deep LLM expertise
Establish the evaluation practice for AI-assisted development at Baseten, and use it to drive investment decisions
Drive adoption through developer experience — good defaults, clear docs, low friction — not mandate
Embed with teams to find where AI genuinely unblocks them, then generalize those wins into platform capabilities
Own the safety layer: permissions, secrets handling, audit trails, cost management