Max Headroom 
The following article was originally published on Tim O’Brien’s Medium page and is being reposted here with the author’s permission.
If you’ve spent any time around AI-assisted software work, you already know the moment when the Scope Creep Kraken first puts a tentacle on the boat.
The project begins with a real goal and, usually, a sensible one. Build the internal tool. Clean up the reporting flow. Add the missing admin screen. Then someone discovers that the model can generate a Swift application in minutes to render this on an iPhone, and the mood in the room changes.
“Why not? We can render this on an iOS application, and it will only take 10 minutes. Go for it. These tools are amazing. Wow.”
That first idea is often genuinely useful. Something that might have taken a week now takes an hour. That is part of what makes the pattern so seductive. It doesn’t begin with incompetence. It begins with tool-driven momentum.
The meeting continues, “Let’s put the entire year’s backlog into the system and see if we can get this all done in a week. Ignore the token spend limits, let’s just get this done.” What was a reasonable weekly release meeting has now set the stage for a rapid expansion in scope, and that’s how the Scope Creep Kraken takes over.
Scope creep is older than AI, of course. Software teams have been haunted by “while we’re at it” long before anybody was pasting stack traces into a chat window. What AI changed was the rate of growth. In the old version of this problem, extra scope still had to fight its way through staffing constraints. Somebody had to build the feature, debug it, test it, and explain why it belonged. That friction was often the only thing standing between a focused project and an over-extended team.
AI broke that.
Now the extra feature often arrives with a demo attached. “Could we add multi-language support?” Forty-five seconds later, there is a branch. “What about generated documentation?” Sure, why not? “Could the CLI accept natural language commands?” The model appears optimistic, which is enough to make the whole thing sound temporarily reasonable. Each addition looks manageable in isolation. That is how the Kraken works. It does not attack all at once. It wraps around the project one small grip at a time.
Signs the Kraken is already on your boat
- Features appearing without a ticket
- Branches nobody asked for
- Demos replacing design decisions
- “It only took the model 30 seconds.”
The part I keep seeing on teams is not reckless ambition so much as confident improvisation. People are reacting to real capability. They are not wrong to be excited that so much is suddenly possible.
The trouble starts when “we can generate this quickly” quietly replaces “we decided this belongs in the project.” Those are not the same sentence.
For a while, the Kraken even looks helpful. Output goes up. Screens appear. Branches multiply. People feel productive, and sometimes they really are productive in the narrow local sense. What gets hidden in that burst of visible progress is integration cost. Every tentacle has to be tested with every other tentacle. Every generated convenience becomes a maintenance obligation. Every small addition pulls the project a little farther from the problem it originally set out to solve.
The product manager might chime in, “A mobile application? I didn’t ask for that, but I guess it’s good. We’ll see. Who’s going to review this with the customer?”
That is usually when the team realizes the Kraken is already on the boat. The original sponsor asked for a hammer and is now watching a Swiss Army knife unfold in real time, with several blades no one asked for and at least one that does not seem to fold back in properly.
AI also makes it dangerously easy to confuse demonstrations with decisions.
The useful response is not to become suspicious of every experiment. Some of the first tentacles are worth keeping. The response is to put the old discipline back where AI made it easy to remove. Keep a written scope. Treat additions as actual decisions rather than prompt side effects. Ask what each new feature does to testing, documentation, support, and the team’s ability to explain the system six months from now. If nobody can answer those questions, the feature is not “done” just because the model produced a convincing draft.
What makes the Scope Creep Kraken a good name is one that teams can use in the moment. Once people can say, “This is another tentacle,” the conversation gets clearer. You are no longer arguing about whether the idea is clever. You are asking whether this is motivated by a requirement or a capability.
AI & ML, Commentary
Radar
