Start With the Problem, Not the Tool
Every week a new model or feature promises to change how you work. Most won't. Here's the filter I use before I let any of them near my workflow — and the three principles that have actually held up.
With so many new AI tools launching every month, it's tempting to try them all. That's the mistake. Case in point: when a major lab ships a flashy new agent framework, it convinces thousands of builders to spend a weekend wiring it up — and most of those integrations are abandoned within a month.
Before adopting anything new, ask three questions: what problem does this solve, who's already using it well, and what would I stop doing if I said yes to it.
01Start with the problem, not the tool
Pick the task that wastes the most of your week first, then go looking for the narrowest tool that fixes it. A general-purpose assistant bolted onto a problem you haven't defined just adds a new interface to babysit.
02Keep a human checkpoint on anything irreversible
Drafting, summarizing, and searching are safe to automate fully. Sending, deleting, and publishing are not — not because the models are unreliable, but because the cost of a silent mistake is higher than the time you saved.
03Review your stack every quarter, not every launch
Reacting to every release keeps you perpetually mid-migration. Set one recurring date to reassess what's actually in your workflow, and ignore the noise in between.
Here's what I'd do next if you're starting from zero:
- Write down the three tasks that ate the most time last week.
- For each one, find the smallest tool that solves it — not the most powerful.
- Give it two weeks before deciding whether it earns a permanent spot.
Resources
- The Prompt Lab — a working library of prompts organized by task, not by model.
- Automation Toolkit — the two automations worth setting up before any others.
- Subscribe to the newsletter — one AI tip a week, nothing else.