Reimagining the team and its roles

Estimated read time: 9 minutes

AI means public service teams are now building in a different medium. In a vibe-coded world, the tempo changes, and that forces a reckoning: we cannot keep the same division of labour and pretend nothing else will alter.

Some roles will shrink. Some will merge. Some will reappear as a different kind of work entirely.

That is not because the work was pointless, but because a chunk of what used to require a person now looks like a capability. Drafting, scaffolding, first-pass analysis, boilerplate, testing, documentation, data schemas, even the first cut of an interface: AI assistance makes those things cheaper. So the question becomes less “how many hands do we have?” and more “where is judgement still needed?”

This is where we might get carried away. Quality working code can be written quickly now but that does not mean the organisation can safely ship everything quickly. The work that remains is less visible: clarity, intent, evidence, operability, and responsibility.

So teams may be smaller, or at least denser. We may end up with the same number of staff across more teams. We may build more in-house, and maintain it properly. We may finally pay down the legacy we have been living around for years, including the unglamorous refactoring of COBOL and FORTRAN estates that still hold up large parts of the state. All of that is in service of improving our services.

And some old boundaries will blur.

Boundaries blur; standards don’t

Do we still need “frontend vs backend vs webops” as hard categories if more of the team can work full-stack on small slices? Probably less than we do today. 

Do we still need “QA” as a separate gate at the end? Much less.

Do we still need quality, testing, accessibility assurance, performance validation, security review, privacy practice? More than ever.

“QA” doesn’t go away but the shape of it changes. Manual regression-as-a-phase becomes harder to justify when the team can generate tests, run them continuously, and ship in smaller slices. But someone still has to care that the service behaves under real conditions, that accessibility holds, that failure modes are humane, that changes don’t create unfairness, and that metrics are not lying. If we remove the people who hold that line, we will not go faster; we will simply push the costs into incidents, complaints, and frontline burden.

This is why the multidisciplinary team remains non-negotiable. If anything, it becomes more important, because at higher speed you can institutionalise blind spots: you can move faster with fewer hand-offs, but you cannot move faster with fewer perspectives.

Two rings that behave as one team

A useful way to make this concrete is to think in two rings that behave as one team:

The core delivery ring

Product, design, research, content and engineering: accountable for user needs, interaction and content design, code and tests, measurement and iteration.

The institutional ring (policy‑in‑the‑loop)

Policy, legal, commercial, assurance, security, data protection, finance, operations, communications, risk and ethics: not as periodic sign-off, but as partners close enough to shape the work while it is still cheap to change.

This is not an org chart. It is a practical claim about what it takes to move quickly without losing your footing. In a vibe-coded world, each discipline still exists, but the cadence changes. Less waiting. Less theatre. More working evidence. More responsibility.

I’ve had a go at naming the roles that have to be present in the system. Not because every team needs every job title in the stand-up, but because every team needs access to these perspectives, at the right moments, in a way that is fast enough to be real.

Roles and accountabilities in a vibe-coded world

This is not an org chart. It’s a statement of work that must be covered if we are to move quickly without becoming reckless. Some of these accountabilities sit in the core delivery ring. Some sit in the institutional ring. Some are best held in enabling teams that create paved roads for everyone else.

Each role description is intentionally short: what this discipline exists to protect; what changes when building is cheap; what “good” looks like when you can ship in days.

Product
  • Where it sits: Core ring
  • Still accountable for: user need, outcome focus, sequencing, and stopping work that doesn’t earn its place.
  • In a vibe-coded world: less slideware; more working code; more decisions made in the presence of evidence.
  • Signs it’s working: a small set of outcomes that stay stable; clear hypotheses and stop rules; learning captured and reused; fewer late surprises.
Engineering
  • Where it sits: Core ring + Enabling (for paved roads)
  • Still accountable for: reliability, security, accessibility, operability, and maintainability.
  • In a vibe-coded world: less typing; more review and verification; more defaults baked into pipelines; more focus on the parts that fail in the real world.
  • Signs it’s working: small reversible changes; tests and checks run by default; preview environments and feature flags are normal; the service can be operated calmly.
Design
  • Where it sits: Core ring
  • Still accountable for: usability, coherence, accessibility as experience, not compliance.
  • In a vibe-coded world: faster iteration in real components; more “showing” earlier; less debating hypotheticals.
  • Signs it’s working: awkward states tested; accessibility holds under change; drop-offs and confusion reduce; design decisions cite evidence not taste.
Content
  • Where it sits: Core ring
  • Still accountable for: clarity, accuracy, inclusion, and words that match the service’s behaviour.
  • In a vibe-coded world: drafting accelerates; editing and truth-checking become more central; comprehension is tested sooner.
  • Signs it’s working: fewer avoidable contacts caused by wording; clearer error messages; consistent tone; fewer “confident but wrong” moments escaping into production.
Research
  • Where it sits: Core ring
  • Still accountable for: ethical, inclusive research; insight you can trust; keeping the team close to reality.
  • In a vibe-coded world: continuous discovery; prototypes used as research artefacts without replacing real users with simulated ones.
  • Signs it’s working: regular user contact; synthesis that changes decisions; fewer untested assumptions reaching live; clear confidence levels on findings.
Delivery
  • Where it sits: Core ring
  • Still accountable for: flow, coordination, team health, and making work visible without bureaucracy.
  • In a vibe-coded world: fewer ceremonies; more cadence; tighter WIP; release choreography; guardrails enforced through defaults.
  • Signs it’s working: pace without heroics; predictable throughput; clear ownership; fewer “urgent” handovers; learning absorbed rather than piled up.
Policy
  • Where it sits: Institutional ring
  • Still accountable for: intent, proportionality, fairness, and explainability.
  • In a vibe-coded world: policy is expressed earlier as runnable rules and worked examples; constraints are co-authored while change is still cheap.
  • Signs it’s working: fewer late blockers; fewer mismatches between policy and service behaviour; clearer traceability; confidence earned through working evidence.
Legal
  • Where it sits: Institutional ring
  • Still accountable for: lawfulness, transparency, equality duties, accountability.
  • In a vibe-coded world: joins the cadence; shapes constraints and notices alongside delivery; keeps DPIA/ROPA live as behaviour changes.
  • Signs it’s working: fewer last-minute rewrites; notices match what the service does; decisions survive scrutiny without reconstruction.
Security
  • Where it sits: Institutional ring + Enabling
  • Still accountable for: protecting people, data and services; proportionate controls; abuse resistance.
  • In a vibe-coded world: controls codified in pipelines; threat models kept living; AI-specific risks named and tested.
  • Signs it’s working: vulnerabilities caught early; secure defaults; rehearsed incident/kill-switch routines; fewer “we’ll harden later” surprises.
Information governance & data protection
  • Where it sits: Institutional ring
  • Still accountable for: lawful basis, minimisation, retention, transparency, auditability.
  • In a vibe-coded world: obligations expressed as checks; tool/model inventories kept current; “no PII in prompts/logs” treated as normal hygiene.
  • Signs it’s working: notices match behaviour; low PII sprawl; DSAR/FOI manageable because records exist; fewer privacy risks discovered late.
Governance & assurance
  • Where it sits: Institutional ring
  • Still accountable for: proportionate assurance; alignment to the Service Standard; accountable decisions.
  • In a vibe-coded world: fewer big gates; more frequent, lighter evidence check-ins on the running thing.
  • Signs it’s working: quick decisions under clear guardrails; stop rules used; issues caught in preview not production; assurance that reduces rework.
Commercial & procurement
  • Where it sits: Institutional ring
  • Still accountable for: value for money, fairness, contestability, sensible exit.
  • In a vibe-coded world: tranche funding against evidence; evaluating working slices not slideware; treating AI tooling as supply chain.
  • Signs it’s working: smaller bets; clean off-ramps; contracts requiring operability and transparency; portability tested, not promised.
Operations & support
  • Where it sits: Institutional ring (close to core cadence)
  • Still accountable for: reliability on a bad day; humane failure modes; frontline reality; incident response.
  • In a vibe-coded world: co-design in preview; rehearsing failure; pacing change using error budgets and real signals.
  • Signs it’s working: fewer escalations; faster recovery; clear runbooks; staff trust; fewer workarounds becoming the service.
Data & performance analysis
  • Where it sits: Institutional ring / Enabling
  • Still accountable for: measurement integrity; baselines; honest interpretation; guarding against gaming.
  • In a vibe-coded world: near-real-time signals; experiment design; confidence stated; metrics treated as instruments, not weapons.
  • Signs it’s working: shared definitions; decisions improve rather than thrash; numbers used to learn, not to posture.
Communications
  • Where it sits: Institutional ring
  • Still accountable for: clarity about what changed and why; trust; accessible comms.
  • In a vibe-coded world: comms aligned with progressive rollout; tight loops with support and ops; quick, humane incident comms.
  • Signs it’s working: fewer “what happened?” contacts; fewer surprises; consistent language across channels; trust maintained through change.
Risk & ethics
  • Where it sits: Institutional ring
  • Still accountable for: proportionality; fairness; accountable choices; seeing harm early.
  • In a vibe-coded world: risk stays live and tied to flags/measures; targeted checks where stakes are higher; provenance and decision logs maintained.
  • Signs it’s working: clear stop rules; harms prevented rather than explained afterwards; fewer “the model said it was fine” moments.

Download the full paper as a PDF.