Read this before you build
Projects are things you make. Guides are the things worth knowing before you choose which one — what the words actually mean, which tool does which job, and where the traps are.
Foundations
What people actually mean when they say "bot"
"Bot" covers eight genuinely different kinds of software, from a fifty-line scraper to a multi-agent system. Grouped by the job they do, with the honest cost and difficulty of each.
Bots are agents that stopped needing to think
A bot isn't the primitive thing agents replaced — it's the part of an agent whose decision has already been made, running reliably at zero marginal cost. Where to spend a model, and where not to.
Scheduling & cost
Two ways to make AI do something every morning
One route has a cost per run and the other doesn't. Invisible at once a day, decisive at a thousand — with the arithmetic, the task ceilings, and how to choose.
The code route: use AI once, then run it free forever
Have a model write the script, then let a free scheduler run it. No task limits, no per-run cost, no drift — plus the three Actions traps that catch everyone.
ChatGPT scheduled tasks: what they do, and where the ceiling is
Task limits by plan, event triggers, and the memory gap that makes change detection the one thing this shape is worst at.
Claude scheduled tasks: recurring work with your tools attached
Built for recurring work rather than reminders, with connectors to the systems the work lives in — and one caveat about local folders.
Gemini scheduled actions: ten slots, and one detail that matters enormously
Responses are prepared in advance, so anything time-sensitive arrives stale. What that rules out, and what it's genuinely good for.
Scraping
Running things
Strategy
Reference
Every bot library, grouped by the job it does
Fetching, parsing, driving a browser, crawling, scheduling, talking to chat platforms, storing results, staying unblocked. What to reach for in each group, and what to ignore.
Where to get the data: stocks, weather, books, maps and more
Check for an API before you write a scraper. Free tiers, real limits and licence traps across finance, weather, books, maps, government data and a dozen more domains.
Know what you want to build?
Answer four questions and we'll order the whole project library around it.