Senior software engineer, 10+ years of production web products in TypeScript. Since March 2026 I work full-time on agent tooling: the runtime, evals, and analytics around coding agents.
Now
- Coding-agent runtime — a TypeScript agent loop with tool calling, Zod-validated tool schemas, sandboxed test runs, cost and call limits, and fixture-based bug-fixing evals. Prototype.
- Evals — a harness that changes one variable per preset (model, backend, single-pass vs map-reduce) and has LLM judges rank the reports. Every claim is backed by a quote the server verifies against the source.
- Agent analytics — a CLI that collects local Claude Code and Codex histories, prices the usage, flags risky tool calls, and renders an HTML report.
- bb plugins — LLM thread titles, a doctor for stuck threads, idea-trial replays that compare tokens and time, a context-window analyzer. I use them every day.
Before
- Smartling (2021–2026) — frontend for a localization platform: the linguist workbench and the admin dashboards. Led a plugin-based multilingual editor built on Slate that replaced the legacy DraftJS editor for thousands of linguists. Introduced Playwright E2E for it.
- AppFollow (2017–2021) — built the company's first React SPA from scratch; grew from junior to senior on the same codebase.
Stack — TypeScript, Node.js, React · Claude / OpenAI / OpenRouter APIs, tool calling, structured outputs · Claude Code, Codex, pi.dev · Playwright · PostgreSQL, SQLite
Open to applied AI and agent-tooling roles, remote or relocation. 📫 work@alecoff.com · LinkedIn