Observe → Co-design → Ship — the Tinkso method

TL;DR. Tinkso's signature method, made public. Observe the work in week one (no tools, just looking). Co-design prompts with the operator in weeks two to four (operator drives, Tinkso engineers tune). Ship in weeks five to six (skills, rollout, ROI checkpoint). The order is non-negotiable; the durations compress on accelerated engagements but the sequence does not.

The problem this method solves#

Tool-first AI rollouts. Most companies — internal IT, Big-Four partners, generic AI consultancies — start with the tool and look for places to use it. The result is skills nobody asked for and adoption curves that flatline at week six. The fix is to start with the work, not with the tool.

The three beats#

Observe — week 1. Watch the work. Two shadow sessions per role, around 90 minutes each. Pain/gain map per role. No tools, no prompts — just looking. Output: shadow notes, pain/gain maps, baseline time data.

Co-design — weeks 2–4. Sit with the operator and write the prompts together. The operator drives; Tinkso engineers tune. Iterate on real data. Save what works to CLAUDE.md. Output: 3–5 production prompts and 1–2 candidate skills.

Ship — weeks 5–6. Convert the best prompts into skills. Roll out to the next 3–5 teammates. Measure. Decide whether to scale.

Why "observe" comes first (and is non-negotiable)#

Three concrete reasons:

• Specificity beats generality. A prompt designed for the actual close process beats a prompt designed for month-end close in general every single time.
• Trust is built by being seen. The team that gets shadowed without judgment opens up about what is actually slow. The team that does not get shadowed defends the process they are supposed to follow.
• Edge cases surface only in the work. No interview surfaces the half-day-of-Friday work that does not fit the official process. Shadowing surfaces it on day two.

Why "co-design" is not "we build it for you"#

• The operator is in the room when the prompt is written.
• The operator says when a constraint matters and when it does not.
• The operator owns the prompt the moment it is saved to CLAUDE.md.
• Tinkso writes the technical pieces (skill packaging, connector wiring); the operator owns the meaning.

The co-design distinction is what makes the prompts survive after we leave. A prompt the operator helped write is a prompt the operator will defend; a prompt handed cold to the operator is a prompt the operator will quietly route around.

What "ship" actually requires#

• Prompts converted to skills (when reuse is real).
• Documentation in CLAUDE.md.
• A six-week ROI checkpoint with the sponsor.
• A written decision: scale, hold, or stop.

The written decision is the artifact. Even stop counts as a successful ship — you stopped before scaling the wrong thing.

How long each beat takes#

At standard pace:

• Observe: 1 week, concentrated.
• Co-design: 2–3 weeks.
• Ship: 1–2 weeks.
• Total: 4–6 weeks per pilot.

Not negotiable: the order. Even on accelerated engagements, we never skip the Observe beat. We compress it; we do not skip it.

What the method looks like across multi-function rollouts#

After the first pilot ships, the next pilot starts with a shorter Observe (because we know the org now), a Co-design that reuses patterns, and a faster Ship. By the third function, the cycle is roughly three weeks. The compounding is real but it is not free — every function still gets its own Observe beat, just shorter.

When this method is overkill#

Honest section: if the work is purely tool installation ("install MCP connectors for our SaaS apps"), no method is needed beyond standard project management. The method is the thing for behaviour-change rollouts — where the people change as much as the tooling. If your project is genuinely just an IT install, run it as an IT install.

Tinkso's take#

We named the method to make it falsifiable. If you hire us and we skip the Observe beat, we did not run our method — and you should call that out. If you adopt the method without us, we will cheer.

Try this#

Before your next AI rollout decision, run the Observe beat in miniature: two 90-minute shadow sessions with the highest-leverage role on the team. Do nothing else for that week. The pilot you would have run before will be wrong; the pilot you design after will be the one worth shipping.