Agentic AI for Teams

This framework is for delivery leaders and engineering organisations navigating the adoption of Agentic AI. It treats adoption as an organisational-capability question rather than a tooling question, grounded in XP practices, continuous delivery, spec-driven development, Team Topologies, and organisational behaviour theory.

Agentic AI for Teams

Adopting agents is a delivery capability problem first

AI amplifies the engineering culture you already have

With strong delivery practices, agents accelerate good work.

Without them, agents accelerate entropy: more code, produced faster, understood less.

The tools are the least interesting part.

profound.fyi

"Agentic AI" is three different problems

  • Developer tooling agents speed up code production. They need tests, CI/CD, and clear boundaries.
  • Business process agents speed up operations. They need defined processes, data ownership, and explicit rules.
  • Customer-facing autonomous systems act for the organisation. They carry the most risk.

This framework focuses on the first.

profound.fyi

Adoption happens at three levels that do not move together

  • Individuals progress through eleven stages, from copying out of a chat window to building software factories.
  • Teams choose between four modes for each piece of work. They do not climb them.
  • Organisations provide the infrastructure and governance that limit what teams can do.

A capable developer in an unready organisation still stalls.

profound.fyi

Most people stall between supervised and autonomous

Moving from approving every step to reviewing output needs three things:

  1. Tests that catch agent errors before production
  2. CI/CD that gates deployment on those tests
  3. Observability that detects what the tests miss

This is where to focus enablement effort.

profound.fyi

Sequence adoption deliberately

  1. Clarify Map the value stream and set a baseline
  2. Harden Build the guardrails that make trust possible
  3. Ease Run bounded experiments and measure flow
  4. Accelerate Expand autonomy as confidence grows

Use AI to get to good before using AI to go fast.

profound.fyi

Rollout and readiness are separate

An organisation can give everyone an agent subscription while nobody has test coverage.

  • Rollout: try, scale, optimise
  • Readiness: clarify, harden, ease, accelerate

Most adoption problems live in the gap between the two.

profound.fyi

Every role changes, and nobody has the answer yet

  • Product: specification quality becomes the bottleneck
  • Senior engineers: value shifts from implementation to judgement
  • Junior engineers: the learning pathway needs redesigning
  • Quality: moves from a separate check to an enabling function
  • Design: shifts from making things look right to defining what right looks like
profound.fyi

Agentic AI exposes your real management philosophy

The transparency that makes agents auditable also makes human work measurable.

Organisations that trust their people use it to learn. Organisations that only claim to trust them use it to watch.

Governance without engineering foundations is performative.

profound.fyi

Six principles for adoption

  1. Solve the sharpest constraint
  2. Own your interfaces, not your implementations
  3. Structure for flow, not for now
  4. Sense and classify, do not predict
  5. Distribute decisions to where the knowledge is
  6. Adopt as an experiment, not a programme
profound.fyi

Read the full framework

profound.fyi/playbooks/agentic-ai-for-teams

All slides

The framework is structured in nine parts:

  1. The Core Thesis - why AI is an accelerant, not a transformation tool.
  2. Scoping the Problem - the three types of Agentic AI and the tooling-adoption misconception.
  3. Maturity Models - individual stages, team modes and practices, and organisational readiness, with Theory of Constraints as the mechanism behind the critical stall and an explicit bounding relationship between team capability and organisational infrastructure.
  4. The Adoption Sequence - clarify, harden, ease, accelerate, plus a remediation sequence for post-chaos recovery.
  5. Role Positions - product, testing, design, learning, with spec-driven development as the unifying frame.
  6. The Management Question - Theory X/Y as a forcing function; governance requires engineering foundations.
  7. Six Principles for Adoption - constraint, interfaces, flow, sensing, decisions, experiments.
  8. Open Questions - team topology, tacit knowledge, measurement, and board-level risk communication.
  9. Acknowledgements and References.

Elements of the framework are hypotheses rather than proven positions. They are marked as such throughout.

The Artefacts section contains the operational tools that put the framework into action: the Underwriting Pack for board-facing risk communication, the Recovery Playbook for engagements that begin after adoption has already gone wrong, and further artefacts as they develop.

Licensed under CC BY-SA 4.0.