Decision Fatigue and Tool Selection

You have access to more AI tools than any human can meaningfully evaluate. There are at least 40 credible text generators, a dozen image models, eight or nine code assistants, and a growing shelf of automation platforms — and that's before you count the wrappers, the integrations, and the "AI-powered" features getting bolted onto every SaaS product you already use. The conventional wisdom says this abundance is a good thing. More options mean better choices. The research says the opposite.

The Paradox That Applies to Everything

In 2000, Sheena Iyengar and Mark Lepper ran an experiment at a grocery store that became one of the most cited studies in behavioral psychology. They set up a jam tasting booth. On some days, the booth offered 24 varieties of jam. On other days, it offered 6. The booth with 24 varieties attracted more people. The booth with 6 varieties sold ten times more jam. [VERIFY: exact multiplier from Iyengar and Lepper jam study — commonly cited as 10x but may be approximate]

The finding — replicated across dozens of domains since — is that more options increase interest but decrease action. When faced with too many choices, people either choose nothing, choose randomly, or choose and then regret the choice. This is the paradox of choice, and it applies to AI tools with a precision that would be funny if it weren't costing people so much time.

Barry Schwartz expanded this into a broader framework in his 2004 book. He distinguished between "maximizers" — people who try to find the absolute best option — and "satisficers" — people who pick the first option that meets their criteria. Maximizers consistently report lower satisfaction with their choices, even when their choices are objectively better. The act of exhaustive comparison poisons the experience of the thing you chose. You picked ChatGPT, but you keep wondering about Claude. You picked Claude, but you keep checking what Gemini shipped. The comparison never ends because the market never stops moving.

How This Plays Out in AI Tools

The AI tool landscape is designed — not intentionally, but structurally — to maximize decision fatigue. New models ship monthly. Each release comes with benchmarks that suggest the new thing is better than the old thing. Twitter threads compare outputs side by side. YouTube reviewers run the same prompt through five tools and declare a winner based on a sample size of one. The entire information ecosystem around AI tools is optimized for comparison, and comparison is the engine of decision fatigue.

Here's what happens in practice. You're using Claude for writing assistance. It works. Your workflow is functional. Then Gemini 2.5 drops with a new context window and you see a tweet showing it handling a task that Claude fumbled. You spend 45 minutes testing Gemini on your use case. The results are mixed — better in some ways, worse in others. You can't tell if the differences are meaningful or just noise. You don't switch, but you don't fully commit to Claude either. You're in the middle now, where every session starts with a micro-decision: which tool should I use for this? That micro-decision costs you energy every single time, and the energy comes from the same finite pool you need for the actual work.

Roy Baumeister's ego depletion research — while debated in its strongest form — points to something most people recognize from experience: decisions use up a resource, and that resource is shared with self-control, focus, and creative thinking. [VERIFY: current status of ego depletion replication debate] Every decision about which tool to use is a decision not spent on the work itself. The person who has already decided — who committed to a toolset and stopped evaluating — walks into each work session with a full tank. The person still shopping walks in with the tank half empty.

The Evaluation Trap

There's a specific version of this that's endemic to the AI community, and it deserves its own name: the evaluation trap. It looks like diligence. It feels like research. It's actually procrastination wearing a lab coat.

The evaluation trap works like this. You decide you need an AI code assistant. You've heard good things about Cursor, Windsurf, Cline, GitHub Copilot, and Aider. The responsible thing to do is test all of them before committing. So you install Cursor and spend a day with it. Then Windsurf. Then you read comparison posts on Reddit. Then you realize you tested Cursor on a simple task and Windsurf on a hard one, so the comparison isn't fair. You re-test. A week has passed. You have opinions about five code assistants and you've shipped no code.

The trap is that evaluation feels productive because it's intellectually engaging. You're learning things. You're forming opinions. You're developing expertise about the landscape. All of that is real — and none of it produces output. The person who installed Cursor on day one and spent the whole week writing code in it has shipped more than you have, and they probably know Cursor better than you know any of the five tools you tested.

This isn't an argument against evaluation. You should test tools before committing. But the evaluation should be time-boxed and criteria-driven, not open-ended and vibes-based. Define what "good enough" looks like before you start testing. Test against that bar. When something clears it, stop looking. The cost of picking a B+ tool and mastering it is lower than the cost of endlessly searching for the A+ tool and mastering nothing.

The Commitment Dividend

The opposite of decision fatigue is what we might call the commitment dividend. When you commit to a tool — when you stop evaluating and start using — you get back all the energy that was going to comparison. That energy goes to depth instead. You learn the tool's quirks. You build workflows around its strengths. You develop the kind of fluency that only comes from sustained use, and that fluency produces better output than any amount of tool-shopping ever could.

The commitment dividend compounds. In week one, the person who committed and the person who's still evaluating produce roughly similar output. By week four, the committed person has built habits, shortcuts, and intuitions that the evaluator doesn't have. By week twelve, the gap is enormous. The committed person is operating at a level that the evaluator can't reach because they've never stayed in one place long enough.

This is counterintuitive because it seems like the evaluator should have an advantage. They've seen more tools. They have more information. But information about tools is not the same as skill with tools, and in practice, skill wins. The person who knows one tool deeply will outperform the person who knows five tools superficially — not sometimes, but consistently, and the margin grows over time.

The Quarterly Evaluation Model

The hex framework handles this with a specific cadence: evaluate quarterly, not continuously. Four times a year, you set aside time to check the landscape. You look at what shipped. You test anything that seems like a genuine improvement over your current setup. If something clears the bar, you swap it in and commit for the next 90 days. If nothing clears the bar, you keep what you have.

This model works because it separates the evaluation mindset from the production mindset. During the quarter, you're in production mode — using your tools, building depth, producing output. During the evaluation window, you're in research mode — testing, comparing, deciding. The two modes don't coexist well, and trying to maintain both simultaneously is how you end up in the permanent evaluation trap.

Ninety days is long enough to develop real fluency and short enough to avoid getting locked into something that's been genuinely leapfrogged. It's a rhythm that respects both the speed of the AI landscape and the speed of human learning. The market moves fast, but your brain doesn't — and the constraint that matters is the one inside your skull, not the one on the release calendar.

What Decision Fatigue Actually Costs

The cost of decision fatigue isn't just time, though the time cost is real. It's quality. When you're fatigued from decisions, your subsequent decisions get worse. This is why judges grant more paroles after lunch than before — not because they're more sympathetic on a full stomach, but because their decision-making capacity has been replenished. [VERIFY: Danziger parole study — this has been debated in recent literature]

Apply this to a working session. If you spend the first 30 minutes of your session deciding which tools to use, comparing outputs, and second-guessing your choices, you walk into the actual work with depleted decision-making capacity. The writing is less sharp. The prompts are less precise. The editorial judgment — the part that separates good output from mediocre output — is running on fumes. You've spent your best cognitive resources on logistics and left the work itself to run on whatever's left.

The hex constraint is, among other things, a decision-fatigue firewall. By capping your tools at six, you cap the number of decisions you need to make about tools. By committing to those six for 90 days at a time, you eliminate the daily micro-decisions entirely. The tool question is settled. The only decisions left are about the work — which is where your decision-making energy was supposed to go in the first place.

This article is part of the Hex Proof series at CustomClanker.

Related reading: The Cognitive Cost of Tool Switching, Time Audit — Managing Tools vs. Doing Work, The Mastery Curve