Case · AI transformation · B2B SaaS · 2025–2026
Changing the organisation so the agents pay off.
In spring 2026 a B2B SaaS company with about 100 engineers set out to deliver much more, much faster with AI agents. The bet: the organisation has to change first, or the agents won't pay off. Here is what changed, what moved, and where work still waits.
- 34 → 16
- days, median cycle time of a product epic
- 55 → 37
- days, the slowest sixth of epics (85th percentile)
- 43% → 78%
- of product epics inside the 30-day target
- 2 → ≈52
- AI-native engineers out of ~100, April → July 2026
Where we started
Work waited at every hand-off.
From components to the whole product
We tracked the organisation on the Org Topologies map: how much of the product a unit owns (rows) against how cross-functional the unit is (columns).
- Mid 2024 · Y2 / A2. Teams own components. Work waits at every hand-off between teams.
- 2025 · A2 → B2. A cluster owns a business process end to end. Work waits between roles inside the cluster.
- 2026 · B3. Pods of 1–3 own a business area end to end. Work waits outside the pod.
- Next · C. The company's top bets borrow pods for a while.
Our path on the Org Topologies map
What changed
Three moves, each one structural.
Every engineer DRI-ready. Not so that everyone works alone, but so that no pod depends on one person.
12025 · Ownership
A cluster owns a business process
- A pilot cluster owns a whole business process and is renamed after it
- One working standard: WIP limits, vertical slices, review turnaround, 3 Amigos
- Teams assess themselves against the standard practices agreed with leadership
- Feature Lead experiment: one engineer owns a feature end to end
Work then waited between roles inside the cluster: frontend, backend, QA.
22026 · Unit, roles, DRIs
Pods of 1–3, one engineering role
Org design
- Teams of 6–8 → pods of 1–3 with agents
- Frontend + backend + QA → one engineering role
- Team leads → DRIs; every engineer DRI-ready
- The DRI joins every idea once its validation starts
Enablers
- Codebase split into domains with public interfaces
- Agent context written for each domain
- Two months of capacity on foundations
- Hiring open to engineers from any stack
Work now waits outside the pod: cross-domain design and the release gate.
32026 · Bet staffing
Company bets staffed first
- One company list of bets, staffed first
- The top bets pull pods from anywhere in the organisation
- Pods keep owning their business area; the bets borrow them for a while
Less waiting for that bet. Open question: the return date.
What moved
Product work got faster.
Product epic cycle time, days
We measured output, not outcome. On purpose, for now.
Quality held while volume doubled
Not in code review: time per change stayed flat while the number of changes almost doubled.
AI fluency
Half the organisation went AI-native. Few teams orchestrate.
Pods can climb six rungs on their own. The top two need company-wide platforms, guardrails and adaptation.
AI-native engineers (95%+ of merged work AI-assisted), out of ~100
Weekly measurement from merged changes, April → September 2026.
- T8 · Agents in productionA few individualsWhole org
- T7 · Fleet operations Whole org
- T6 · OrchestrationA few teamsDomain
- T5 · Checkpointing and hardening Domain
- T4 · Loop until doneMost of usPod
- T3 · Context management Domain
- T2 · Shaping and slicingEveryonePod
- T1 · Professional prompting One person
Bar = how many of us work at that rung (our estimate)Right = who owns the rung
Side effects
What else the agents unlocked.
Beyond a narrow hiring funnel
Strong engineers from any stack. Agents carry the language learning curve.
Multi-specialist engineers
One engineering role across the stack. Agents fill the gaps.
Quality shifts left
Engineers own testing. AI reviews specs and changes.
Engineers in discovery
Engineers join discovery much earlier, not only delivery. Every engineer can.
Rapid prototyping
Clickable prototypes inside the real product, before anyone decides to build it.
Cross-domain contribution
Another domain's code is easy to work in. Approval is now the wait.
Change the organisation, not the tools
The tools made it possible. Organising around them made it pay off.
Readiness: grow owners, not just AI users
Product thinking is everyone's job, not only the former team leads'. Who takes the role is the pod's call.
Gates: choose them, and revisit them
Every mandatory gate is a choice. Give it an owner, a visible turnaround and a regular “still worth it?”
Foundation: check the ground is agent-ready
Domains with clear boundaries, context the agents can read. Assess it before you scale the agents.
Where does work wait in your organisation?
Book a short call, or start by email if you'd rather write first. Tell me what you're seeing, and I'll tell you whether I'm the right person to help.