When You Can't Stop Prompting
Daria Golub
AI can make work faster, but it often increases expectations and mental strain instead of freeing up time. It can be hard to stop when the work feels exciting and there are no natural pauses. Set clear stop-and-break rules, and write down tasks and plans to ease the mental load.
19 Oct 2026
The researchers identified a cycle they named workload creep: AI accelerates tasks, which raises expectations for speed. Higher speed increases reliance on AI. Greater reliance widens the scope of what people attempt. Wider scope expands the volume and density of work further. There’s no natural stopping point.
The speed feels good — until it doesn’t. On Hacker News, a developer posted: “I used to code for nearly 14 hours a day on weekends, and before AI code agents existed, this felt manageable. However, now that I’m building products at breakneck speed with AI assistance, I’m experiencing severe cognitive overload combined with serious mental burnout.” The pace that AI enables doesn’t feel unsustainable in the moment. It feels exciting. The crash comes later.
Working with AI means holding your intent in working memory while simultaneously evaluating whether the agent’s output matches that intent. You’re building and auditing at the same time. That dual-track processing is cognitively expensive in a way that pure coding never was.
When you’re at capacity, every unmade decision costs you. Which Slack message to answer first. Whether to review that AI-generated PR now or after lunch. When to stop sending prompts to your coding agent. Each micro-decision draws from the same cognitive pool you need for architecture work and complex problem solving.
Implementation intentions are “if-then” rules you commit to in advance, so you don’t have to decide in the moment when your working memory is already full (Gollwitzer, 1999). A meta-analysis across 642 tests confirmed they reliably reduce cognitive load and improve follow-through, even under high pressure (Sheeran, Listrom & Gollwitzer, 2025). They work because they shift decision-making from being effortful to automatic.
Sit down once, during a calm moment, and write specific rules for your highest-friction decision points:
“If two agents are already running, the next idea goes straight to my backlog.” → sequencing
“If I received an AI-generated output, I’ll put it into sequence and review later” → sequencing
“If I work for 1 hour straight, I take a break” → pauses
“If it is 9:30pm, then I close the laptop. Even if the agent is mid-task.” → pauses
Structured Cognitive Offloading
Your working memory holds roughly four items at once (Cowan, 2001). Every unresolved concern — the bug you haven’t logged, the output you are waiting from the agent, the PR you are reviewing — occupies one of those slots. When the slots are full, your ability to do complex work degrades. You feel foggy, reactive, unable to focus.
Cognitive offloading means deliberately moving information out of your head and into external systems, so your working memory is freed for the work that actually requires it (Risko & Gilbert, 2016).