Estimated reading time: 6 minutes
Key takeaways:
- An engineering strategy can be correct and still become wrong as AI changes delivery.
- Match the commitment to the evidence behind it and review long bets often.
- Build in signals that show when assumptions fail.
An engineering leader approves a plan to hire five engineers, invest in an internal AI platform, and reorganize two teams around faster software delivery. The business case is straightforward: AI-assisted development should increase engineering capacity, allowing the organization to deliver more without growing headcount at the same rate.
Six months later, developers are producing more code, but delivery is not moving faster. Pull requests take longer to review, CI queues are growing, and security teams are seeing more changes that need verification.
Nothing necessarily went wrong with the original plan. The problem is that the plan was built around a model of software delivery that AI has changed. More than simply adding another tool to the development stack, AI s changing the economics and constraints of software delivery faster than many planning processes are designed to respond.
The question for leaders is what happens after the strategy is approved. How does the organization know when the conditions behind a major engineering decision have changed enough to require a different decision?
Your inbox, upgraded.
Receive weekly engineering insights to level up your leadership approach.
A strategy can be correct and still become wrong
Hiring five engineers, reorganizing teams, and investing in AI tooling could all be sensible decisions. However, the limiting factor can shift from how quickly can engineers build this to how quickly can the organization verify, integrate, and operate it?
That creates a second-order problem: AI can improve the metric used to justify a strategy while making the overall delivery system more constrained. The organization improves local productivity without necessarily improving system-level delivery.
LeadDev’s AI Impact Report 2026 provides a useful signal: 68% of organizations identify a lack of clear metrics as their biggest AI challenge, while only 31% say they are effectively measuring AI’s impact. This is why the question cannot simply be: “are we using AI?” It has to be: “what changed in the system because we started using AI?”
A September 2026 CloudBees survey of 213 enterprise technology leaders found that AI now writes or assists with 61% of code in the average enterprise. Yet 57% identified reviewing, testing, and deploying code as the delivery bottleneck, compared with 35% who identified writing it.
For engineering leaders, the strategic implication is straightforward: increasing code-generation capacity without increasing the organization’s ability to validate that output can move the bottleneck rather than remove it.
When the planning cycle becomes part of the problem
Annual and quarterly planning provide necessary discipline. Budgets, hiring plans, team structures, and major technical investments cannot change every week. AI can introduce changes between those planning points that materially affect the economics of an existing decision.
Gartner’s warning about the durability of AI business cases matters here. In June 2025, Gartner predicted that more than 40% of agentic AI projects would be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. The assumptions behind an AI investment can expire faster than the investment cycle.
A long-term commitment does not need a long gap between reviews. A two-year investment can have a quarterly review horizon. A five-year architectural direction can contain six-month checkpoints.
For each significant strategic commitment, leaders should be able to answer:
- What are we deciding?
- What must remain true for this decision to make sense?
- What signals will show whether those conditions still hold?
- What threshold would cause us to reconsider?
- When will we deliberately review the evidence?
The purpose is not to turn strategy into a spreadsheet but to make the organization’s decision logic visible.
More like this
Match the commitment to the decision
Not every strategic decision deserves the same level of commitment. A major platform migration, infrastructure investment, or organizational restructuring can take years to reverse. A limited AI pilot or team-level workflow experiment can be changed much more easily.
The principle is straightforward: match the durability of the commitment to the durability of the evidence behind it. Amazon formalized a version of this in Jeff Bezos’ 2015 shareholder letter, distinguishing irreversible “one-way door” decisions from reversible “two-way door” decisions.
AI-era engineering strategy needs the same classification instinct. The failure mode is applying irreversible-decision discipline to a reversible experiment, or committing heavily to an assumption that has not earned that level of commitment.
Build a strategy that can detect when conditions change
No engineering leader can reliably know which models, agent architectures, or developer tools will dominate several years from now. A strategy built around those predictions simply creates another dependency on a forecast. A stronger approach is to make the strategy capable of detecting when its underlying case is no longer holding.
One practical mechanism is an assumption register for major strategic bets: a small set of statements that need to remain true for each decision to make sense. Each should have an observable signal, an owner, and a review trigger.
| Strategic assumption | Signal | Trigger |
| AI-assisted coding increases delivery capacity | Lead time, throughput, review time | Reassess if review capacity grows slower than implementation |
| Internal AI infrastructure remains economically justified | Cost, utilization, external alternatives | Reassess if external capability changes the economics |
| Quality controls scale with AI-generated changes | Defects, review findings, test maintenance | Reassess if verification becomes the dominant constraint |
Give the signal a path to a decision
An assumption register is useless if nobody is responsible for acting on what it reveals. The leader responsible for a strategic bet should own its decision threshold, and engineering leadership defines the operational signals that matter. The people who see the signals and the people who can act on them need to be connected.
A platform team may know that CI capacity is becoming the constraint. A security team may know that generated changes require a different review process. Staff engineers may see teams independently building overlapping agent infrastructure. Those signals have strategic value only if someone with authority can use them to revisit the underlying decision.
If autonomous coding increases the number of changes an organization can produce but also requires substantial investment in verification, dependency controls, sandboxing, and auditability, those costs belong in the strategy rather than being discovered later as operational overhead.
The organization should know in advance who can trigger a reassessment when review time crosses a threshold, AI infrastructure costs exceed the business case, or agents begin operating across production systems.
Berlin • November 9 & 10, 2026
Close the gap between what leadership expects and what’s actually possible at LeadDev Berlin.
The roadmap needs an expiry mechanism
AI does not eliminate the need for long-term engineering strategy. It changes what a good strategy needs to contain. A conventional roadmap says what the organization intends to build and when.
An AI-era strategy also needs to say why the decision makes sense now, what must remain true for it to keep making sense, and what evidence would cause the organization to reconsider it.
The roadmap does not need to change every quarter. The evidence behind the roadmap needs permission to.
The strongest engineering strategy may therefore not be the one with the most confident five-year plan. It may be the one that makes clear what the organization has committed to, why it made that commitment, how it will know whether the assumptions still hold, and what it will do when the evidence says it is time to change direction.
