AI Delivery Manager
Summary
Manage end-to-end product delivery at a blockchain/staking company by leveraging AI coding agents (Claude Code) to spec, build, test, and ship internal tooling and features — writing PRDs, decomposing work, running parallel agent sessions, and owning the review/validation layer for production output.
What you'll do
Take a business problem and deliver a working product — spec, decomposition, implementation, testing, release
Run multiple parallel agent sessions across projects; manage context, worktrees, and merges
Write PRDs, use cases, and test cases that agents can actually execute against
Define and build the team around this model — decide which roles are actually needed, write the profiles, interview, and onboard.
Own the review layer — decide what gets reviewed by agents, what needs a human, and who that human is. Set the standard for what "validated" means before something reaches production
Pick up projects currently stuck in the backlog and prove they can be delivered without a development team
Extend our existing systems where we've hit a ceiling — BD tooling, data science workflows, internal automation, our meeting/insight memory bank
What we're looking for
Required
Track record of managing people or projects — you're comfortable defining work and holding others to it
Demonstrated agentic delivery: you've shipped something real with Claude Code or an equivalent harness, not just used AI for autocomplete
Strong decomposition and context management — you know how much context an agent needs and how to give it
Broad applied expertise in some domain (trading, fintech, ops, infrastructure, data — we're open). Depth in one narrow stack matters less than range
Fluent working English, including when prompting
Nice to have
Your own SDLC harness or workflow you've refined and reused
Comfort with GitHub as issue tracker and CI surface, automated E2E testing
Experience with LLM observability and prompt analysis
Blockchain or staking context (we run a small surface: staking, cashback calculation, a few light smart contracts)
Not a fit if
Your goal is to go deeper into hands-on coding as a craft
You expect a spec handed to you rather than one you write
You treat agent errors as evidence the approach doesn't work rather than as your problem to solve