The Agentic Enablement Playbook
The harness is the product. Code is the output.
Most AI adoption stops at giving people a chatbot. The Agentic Enablement Playbook is an operating model for going much further: turning a team of specialists into an organization where everyone builds, agents do the execution, and humans own the reasoning. The core idea is the harness: everything an agent needs to work like a seasoned member of the team, from organizational context to reusable skills to automated quality gates.
It grew out of a practice I call Intent Engineering: grounding agents in organizational intent, from values and live OKRs to domain architecture and product philosophy, so they onboard instantly and act with product awareness. The playbook is what happens when that grounding stops being a technique and becomes how the whole team operates.
I developed this playbook from firsthand experience leading teams through the shift to agentic-first delivery: first the teams that build Search and AI experiences, then my peer teams across the Marketplace organization, 8 teams in total. On those teams it has reached the mature stage: agents author nearly every pull request, cycle time has fallen from weeks to hours or days, and every product manager and designer ships code. I also keep building hands-on in personal side projects, end to end through agent orchestration.
Ground the Harness
Before any workflow gets automated, the organization itself becomes context. Architecture, live OKRs, process, protocol, product definitions, and team domain boundaries are codified as agent-readable artifacts. This is the starting point, and it is Intent Engineering: an agent grounded in organizational intent onboards instantly and makes product-aware decisions instead of generic ones.
The harness is a living artifact, not a setup step. OKRs roll over, architecture drifts, protocols evolve. Every incident, every confused agent, and every rejected contribution is a patch. A stale harness is worse than none, because agents act confidently on wrong context.
Map the Human Work
With the harness in place, the team turns the lens on itself: outline the playbooks behind everyday work and reflect honestly on which parts require human reasoning and which parts are clerical. The goal is not automation for its own sake. It is finding where judgment actually lives, then aiming the team’s time at it.
Complex workflows get decomposed into discrete tasks. Decomposition is what makes the next stage possible: a task with clear inputs, outputs, and standards is a task an agent can own.
Codify the Skills
Each mapped workflow becomes a reusable agent skill: an executable playbook any contributor can invoke, carrying the team’s standards with it. What used to live in one senior engineer’s head now lives in infrastructure the whole organization can run.
Skills compound. Every codified workflow lowers the cost of codifying the next one, and the library becomes the team’s real velocity asset.
Build the Gates
Before the doors open, quality has to stop depending on heroics. Evals, tests, CI, policy-as-code, and protected paths make standards enforceable by construction rather than inspection. Review shifts from gatekeeping to sampling.
The gates are what make broad contribution an invitation instead of a hazard. They are also the honest answer to the blast-radius question: contributors from any discipline can ship safely when safety is built into the pipeline, not delegated to vigilance.
Open the Domains
The tipping point. Every team commits to accepting contributions from every discipline: engineers crossing domains, designers and product managers shipping working solutions against the same stack and standards. This is where controlled chaos happens, and where the compounding starts.
One norm makes it work: when a contribution looks wrong or breaks, it is not on the contributor. It is the domain owner’s signal to harden their harness, skills, and gates, treating the failure as a dogfood exercise. Every bad PR becomes better infrastructure. Teams escape service-desk work, and contributors take on fulfilling, complex work end to end, anywhere in the org.
The Maturity Ladder
Stage five is not a switch, it is a progression. Where a team sits on the ladder is measured by how much human shepherding a cross-domain contribution still needs.
Early
- Cross-domain contributions arrive as handoffs. The owning team invests real manual review time and shepherds each one to production-ready state.
- Contribution from outside the team may be limited to technical contributors while the harness and gates harden.
- Failures are frequent teachers. Each one patches the harness.
Mature
- Manual review time approaches zero. Sampling and gates carry quality.
- Non-technical stakeholders ship functioning, production-ready, standards-compliant solutions.
- Time to market is structurally reinvented. The org runs its own internal equivalent of a Replit or Lovable, on its own stack, governed by its own gates.
What Humans Own
The playbook automates execution, not thinking. At maturity, people spend their time where they are irreplaceable.
- Product judgment and taste: what to build, for whom, and why.
- The macro architecture of solutions: systems that scale, boundaries that hold.
- The harness, the skills, and the gates themselves: the machine that builds the product.
How You Know It Is Working
- Manual review time per contribution trends toward zero.
- The share of shipped work crossing domain or discipline lines keeps rising.
- Time to first shipped contribution for a new or non-technical contributor keeps falling.
- Cost per shipped change drops as the skill library and gates compound.
The endpoint is a team whose job is not writing code. It is building the machine that builds the product: the harness, the skills, and the gates. That is how small teams ship like large ones, and how an organization scales through capability instead of headcount.
Want to compare notes on agentic delivery?
I’m always glad to talk with leaders working through the same shift.