Planning cycles have a funny relationship with reality. Every year, teams sit down to map what is possible, what is out of reach, and what requires external help — and most of those judgments are recycled from the last time anyone seriously tested them. The world has changed. The judgment, often, has not.
The Gap Between Capability and Willingness
AI has expanded what people can actually do faster than most of us have revised our sense of what we are willing to attempt. The gap is not primarily technical — it is motivational. The constraints we carry about what is practical, what requires specialized help, what is too ambitious for our team were formed under specific conditions. Those conditions have often changed substantially. The constraints tend to stick anyway.
Call these stale constraint maps: internal judgments about feasibility that made sense when they formed but have not been updated to reflect what is possible now. They show up in 2027 planning as items confidently marked “not this year,” projects shelved because they would need a developer, or capabilities written off because last year’s attempt stalled.
The tricky thing about stale constraint maps is that they are invisible from the inside. They do not feel like outdated beliefs. They feel like realism.
What Actually Updates a Constraint
Reading about AI capabilities does not do it. Watching demonstrations does not do it. What actually updates an internal constraint is finishing something — shipping a real project, however rough, that was previously out of reach.
A Reuters journalist built an AI tutor (a custom tool that adapts to her son’s learning pace) for her son. The finished thing mattered less than what building it did to her sense of what she could take on next. That is the generative loop: completion changes what you are willing to try, which changes what you are capable of completing.
This is why “we should pilot something” is often less useful than “we should finish something.” The pilot that gets deprioritised and quietly shelved teaches nothing. The rough first version that ships — even imperfectly — exposes the information you needed and recalibrates what comes next.
Five Skills That Develop Together
There is a cluster of skills that tend to develop together through real project work, and they are worth naming because they do not show up on any job description or capability framework.
Taste — a clear sense of what you actually want — gives a reason to act before you know everything required. It is the motivation that precedes the plan.
Agency — putting judgment to work before you have complete information — is a learnable disposition, not a personality trait. It develops through the practice of starting with partial knowledge and staying with the problem as the approach evolves.
Focus, at this moment in the practice, is more consequential than it used to be. Sustained attention puts you in contact with the hard parts of a problem — the places where a solution has to actually fit — and AI makes the time you spend there matter more. Cutting what does not serve the core (the Stardew Valley solo developer’s lesson: removing a system that worked was the right call) turns out to be a form of focus, not a compromise.
Resourcefulness — finding your way through difficulty encountered mid-project, rather than avoiding hard parts — is what separates projects that finish from ones that stall when the first approach stops working.
Follow-through is what lets you discover what actually helps the person you had in mind. Not what you predicted would help. What you find out by staying with it to the end.
These five develop together. The work you take on by exercising them expands the work you can take on, which develops the skills further.
The Question Worth Asking Before You Finalise the Plan
Personal software — small, custom tools built precisely for one person’s work or daily life — is becoming increasingly common. AI agents like Meta’s Muse (a personal AI assistant that can carry out tasks across a browser and connected apps) are accelerating this trend, moving into territory that people used to write small apps to handle. Work that used to require a developer, or a complicated app, or just tolerating an imperfect process, is now within reach of the person who understands the problem best.
Before locking in which capabilities your team buys, outsources, or leaves out of scope for 2027, it is worth a pass at the parked projects: the ones ruled out because they would need resources you did not have, skills you had not developed, or effort that seemed disproportionate to the payoff.
Some of those judgments will still hold. But some were formed under conditions that no longer apply — the last time you tried was before AI could take the difficult parts off the critical path.
The audit worth doing is not “what is possible in theory.” It is “which of our constraints are still real, and which are echoes from the last time we attempted something like this?”
Text summarized and optimized using Anthropic’s models and reviewed by a human.