Career framework (agent-era addendum)
What progression means when agents do the typing. Code volume and ticket throughput were always weak proxies for engineering value; once agents produce the volume, they are no proxies at all. This addendum redefines the currency of progression — and deliberately protects the pathway that turns intermediates into the seniors the whole model depends on.
The new currency
- Scope of direction — the size and risk of the work you can be trusted to direct: a task, then a feature, then a product, then a portfolio-level concern. Growth is directing bigger, riskier slices well — measured by outcomes, not by hours at a keyboard.
- Review authority — what your approval means: merge-gate approver, then gate-waiver authority (per the RACI), then authoriser of irreversible actions. Seniority is being trusted with progressively less reversible decisions.
- Capital deposited — the prompts, skills, gotchas, runbooks and playbooks you contributed that made other squads faster — the compounding contribution the metrics framework counts as capital-deposit rate. Evidenced in progression conversations with concrete artefacts; never turned into an individual ranking (the anti-gaming rule holds here too).
The junior pathway
The failure mode to design against: agents now do the work intermediates used to build judgement on. Without a deliberate pathway, we stop making seniors. So the pathway is explicit:
- Supervised directing — intermediates direct real agent work with a senior reviewing their direction, not just the output. Bad briefs get caught and coached, the same way bad code used to be.
- Review apprenticeship — structured practice at reading agent output: shadow reviews, then seeded-defect exercises (from the training curriculum), then second-reviewer duty, graduating to solo merge-gate authority.
- Depth rotations — regular stints below the abstraction: hand-debugging agent output, characterisation tests on legacy, incident work. Judgement comes from contact with how systems actually fail; the rotation guarantees that contact doesn't disappear with the typing.
Specialist tracks
- AI engineering (new track) — prompt and skill engineering, guardrail design, golden-task harnesses, model selection and token economics. A first-class discipline, not a side interest.
- Existing tracks re-anchored — engineering, test, BA, design and support tracks restated in direction / review / capital terms and mapped to SFIA levels of responsibility. SFIA works in our favour here: it has always measured autonomy, influence and complexity rather than output volume.
This is an addendum draft. Mapping to grades and compensation, the SFIA level alignment per role, and formal adoption all need HR and Leadership sign-off before it is used in any performance or progression conversation.
Standards referenced: SFIA.