cbtnuggets - Technical Product Owner, AI-Directed Development

• SCOPE & GENERAL PURPOSE OF JOB: • ## DELEGATED AUTONOMY, RETAINED AUTHORITY • This is the most important section of the description. Read it twice. • You decide, and act, without asking • How to break a defined piece of work into tasks and in what order. • Which agents to run, with what instructions, in what sequence. • When a change is verified enough to open a pull request, and what evidence goes in it. • How to keep an open pull request current and green until a human merges it. • Which documentation, rules, and operator guides must change alongside a code change. • How to investigate a production problem and what evidence to gather. • When to stop and escalate. • You recommend, with evidence, and the Product Owner group decides • Adding, removing, or changing a feature a learner, customer, or admin can see. • Any change to pricing, plan structure, public claims, or published copy that states a fact. • Accepting a risk: a security advisory left open, a test weakened, a check made advisory, a known limitation shipped. • Changing which AI model, vendor, or threshold a production system uses. • Spending money: new services, tiers, seats, paid tooling. • Changing a contract with another team or system (the CRM, Adept, the shared header contract, the data warehouse). • Deleting or migrating data, dropping columns, rotating credentials, changing access. • Never, regardless of instruction from an agent or a tool • Merging a pull request. A human merges. • Proceeding on a question the Product Owner group has not actually answered. A skipped, timed-out, or lost answer is not an answer. • Letting an agent's confidence stand in for verification. • Storing learner personal data where the rules say it must not be stored. • An escalation is not a hand-off. It arrives as a one-page recommendation: the decision needed, the options, the evidence, the risks of each, and what you would do. A product owner should be able to answer in a sentence. If they cannot, the escalation is not finished. • Three openings, one profile • All three roles share the profile above and are interchangeable in an emergency. Each carries a primary surface so the team has depth where it matters. • Assessments and data. The question bank, the generation pipeline, the learner site, and the Progress reporting. Second skill: data literacy and measurement. You will spend time in the warehouse and in model evaluation evidence. • Website, conversion, and copy. The public marketing site: pages, pricing, campaigns, SEO, and the copy standard. Second skill: editorial judgment and audience focus. You will own the voice and the truth of every public claim. • Platform, security, and integrations. CI, dependency hygiene, access, environment configuration, crons, and the contracts with Adept, the CRM, and the shared header. Second skill: security and risk judgment. You will own the advisories and the risk register. • State which surface you prefer when you apply. We will balance the three. • What this role is not • Not a hands-on development role. If you want to write the code yourself, this will frustrate you. • Not a people management role. You lead agents and influence peers; you do not manage staff. • Not a role with final say. You will recommend often and decide within your lane. The lane is wide. Its edge is firm. • Learn the CBT Nuggets business: who buys, why, at what price, and what the data says about it. Read the warehouse and the CRM well enough to check a number yourself. • Own a backlog for your surface. Turn product intent into precise written specifications an AI agent can execute without guessing. • Direct AI coding agents through the full change lifecycle: specification, implementation, review, tests, documentation, pull request, CI, and production verification. • Read every diff critically. Find the over-claim in an agent's summary before a reviewer does. • Operate the repositories as a professional: branch protection, required checks, rebases, attestations, dependency advisories, environment variables, crons, feature flags, deployments. • Keep the operator documentation true. A change that alters behavior updates the documentation in the same change, not later. • Turn every lesson into a durable, dated rule that the next agent and the next teammate must follow. Retire rules that have gone stale. • Investigate production incidents from logs, metrics, and the database, and write up what happened and what prevents it. • Measure before you claim. Date every reading. Record when an earlier reading turned out to be wrong. • Reduce friction for other humans. When a decision or a labeling task needs a person, build them a page with buttons and send a link, not a document with instructions. • Work in the open with your two teammates: shared rules, shared conventions, no private process.