A strategy is built on models and values, and it tells you what to do to optimize for those models and values. It isn't a roadmap or a list of projects. A good one lets any engineer make the right call without escalating.
It differs from a runbook. A runbook is a checklist you run. A strategy is a value statement. Less of a checklist, more of a way to choose. With the right models and values, you can work out the next move yourself, in any position.
A strategy is a set of values
In chess, a bad strategy values pawns more than queens. A good one values queens more than pawns. Neither tells you which move to make. With the right values, you can work out the move yourself, in any position.
Engineering works the same way. Say "a deploy is worth more than a dollar of cost, as long as you stay above the SLO" and the next decision follows from it.
Ship as many changes as you can, as cheaply as you can, as long as reliability stays at or above the SLA.
Good strategy vs bad strategy
| Good | Bad |
|---|---|
| Names the real problem in one sentence | Restates the goal ("be more reliable") |
| Picks a trade-off and says no to things | Tries to do everything |
| A few actions that fit together | A pile of unrelated projects and OKRs |
| Plain words | Jargon |
| Gets revisited when the situation changes | Gets written once and filed |
Strategy vs a runbook
| Strategy | Runbook | |
|---|---|---|
| Question | Where do we spend effort? | How do we do this step? |
| Horizon | Quarters | Minutes to days |
| Aided by | Models and values: equations, charts, what each thing is worth | Discipline, training and docs: checklists, practice |
| Fails when | The model is wrong | A step gets skipped |
| Example | Ship as many changes as the SLO allows | Roll back a bad deploy in under 5 minutes |
Rule:if you're deciding, reach for models and values. If you're doing, reach for discipline and a runbook.
Diagnosis, in this framing, is the runbook: the checklist you run once you know where you are. The takeaways of a strategy are its guiding principles. What you can produce are the coherent actions: what you, or your agent, can make once you understand the models and values and follow the strategy.
The operational documents, the checklists themselves, start at Start here. Tool runbooks are under Runbooks.
Takeaways
These are the guiding principles of the engineering strategy, as it stands. They come out of the models below, not the other way around.
- Reliability is a constraint, not a number to maximize. Meet the SLO, then spend the rest of the budget on change.
- Impact is the downtime people feel: how many people, for how long, how much of the system they need. Cut who sees a bad change, and how long they see it.
- Deploy is not release. Put code on a server without making it visible, and you can ship often and still control exposure.
- Make changes cheap to get wrong. Halve the fix time and you double what you can ship.
- If you can't write the check, you don't know what done means yet.
- Work that ships more changes, or cuts toil, goes first. Priority is changes divided by cost.
What you can produce
Follow the strategy, and understand the models and values under it, and you or your agent should be able to produce these without waiting for a new page from me:
- A runbook for a system you have diagnosed: the checklist you actually run. Start from Start here and the runbooks.
- The desired-state checks for a change: what must be true when the work is done.
- A change policy for a service: what can go out unattended, what needs a human, and what "reversible in 5 minutes" means here.
- An agent brief that uses the same models and values, so the agent can propose a change you would actually accept. AGENTS.md is the starter.
The documents
The engineering strategy will be several documents. Reliability and change management is first. DevOps and agentic tool use are coming.
- Reliability and change management. The models for shipping change under an SLO: error budgets, impact, smaller releases, tests, desired state, and deploy vs release.
- DevOps. Coming. Until it is written, use the runbooks and the notes on toil and elite delivery in the reliability document.
- Agentic tool use. Coming. Until it is written, use Working with Claude, the automation blueprint, and AGENTS.md.