A planning meeting that once produced three hard choices now produces fourteen polished options. Every option has a clean format. Several include tradeoff analysis. Someone has already asked a model to draft the rationale.
The meeting runs smoothly. There is plenty to discuss. Nobody gets stuck searching for something to say.
Then everyone leaves with the same uncertainty about what the team should actually do next.
That is the Infinite Backlog Engine.
AI has made the cost of proposing extremely low. A team can generate plans, requirements, goals, briefs, and roadmap ideas faster than people can read them. Each document looks reasonable on its own. The problem is not that every proposal is bad. The problem is that the organization can produce more plausible proposals than it can evaluate.
The backlog swells. The roadmap becomes a museum of almost-good-enough priorities. The documentation gets better while the product moves no faster.
The Engine feels like productivity because the artifacts arrive quickly. You can point to a growing list of plans and say that work is happening. But a volume of intentions is not progress toward an outcome. The useful question is much less flattering: what actually shipped?
The Engine feeds itself through layers of refinement. Someone asks a model to generate a set of OKRs. A manager produces a more polished version. A senior leader adds strategic framing and competitive references. By the time the goals reach another team, they contain language from several rounds of editing but may not reflect a decision anyone consciously made.
Fluency starts doing the work that reading used to do.
People see a plan that looks evaluated, so evaluating it again feels unnecessary. The wording is confident. The sections are complete. The tradeoffs appear to have been considered. The document creates the impression that judgment has already happened, even when nobody has chosen between the options.
This is how planning replaces deciding.
The cure is not simply to generate fewer documents. It is to make selection explicit. A team needs fewer priorities, and it should treat that limit as part of the design rather than as an unfortunate constraint. Every new priority should displace something else or explain why it does not.
That explanation should be concrete. What changed? What work is coming off the plan? Who made the decision? What will the team stop doing to make room?
A useful rule is that adding work requires naming the work it replaces. If nothing comes off, the team has not reprioritized. It has expanded the promise.
The same standard applies to generated plans. Do not approve a document because it is complete or well written. Name the person who read it, understood it, and accepts responsibility for acting on it. "Adopted by" should mean more than a name in a document footer. It should identify someone who can explain the decision without opening the file.
Meetings need a similar test. If a meeting about priorities cannot produce a single exclusion, it was probably a meeting about the backlog. The backlog is a record of possibilities. Priorities are a record of choices. Confusing the two lets everyone leave with a full list and no commitment.
It also helps to separate ideas from obligations. Keep a parking lot for proposals that are interesting but not selected. Give it a clear status. A parked idea is not a future promise, and it should not quietly accumulate supporting documents until it becomes politically difficult to remove.
AI is useful for exploring options, clarifying a proposal, and exposing questions a team may have missed. It is not a substitute for deciding which option deserves time. That responsibility still belongs to people with enough context to accept the consequences.
The friction that AI removed was not all waste. Some of it was selection. When proposing became cheap, teams needed to rebuild that friction deliberately.
Otherwise the Engine keeps running, producing plans nobody has chosen and priorities nobody can defend.