Why this exists

Most AI content stops at the demo — a working example on the author's machine, screenshotted at the moment it behaved. What's missing is everything after: deploying it, paying for it, and what to do the first time it confidently does the wrong thing in front of someone who matters.

Lets Build a Bot is built around that gap. Every project is a complete build sheet rather than an article: what you'll have at the end, what it costs to build and to run, what you need to know before starting, every step with code you can actually run, the ways it tends to break, and how to make it yours. If a project can't be finished in one sitting, it gets split, because half-finished builds are how people give up.

The thesis

There's a specific argument running underneath all of it. Renting reasoning by the run is fine at small scale and ruinous at large scale, because a model call has a marginal cost and code doesn't. A task you run once a day costs pennies a year either way. The same task a hundred times an hour is the difference between free and a salary.

So the skill worth teaching is knowing which half of a job needs a model and which half doesn't — using AI to work something out once, then capturing that as code which executes forever at zero marginal cost. Pay for judgment; never pay for execution. Almost every project here is shaped that way, which is why so many of them cost nothing to keep running.

The scope is deliberately wide. Software agents are the core, because that's where most people can get a useful result fastest. But a lot of the best ideas are about putting AI into something that already exists — a spreadsheet, an inbox, a website — and some of the most satisfying builds are physical: a camera by the door, a sensor in a plant pot, a box on the counter you can talk to. There's also a track on using AI to write software, which is the thing most people actually do with it every day and almost nobody is taught to do well.

The rules we hold ourselves to

Every project has been built and run — on real hardware, or a real account, with real data. Code blocks are complete enough to run; there are no "rest of implementation" gaps. Running costs are stated honestly, including the projects where the honest answer is that it isn't worth it yet. Hardware parts lists include rough prices and flag what can be substituted. Where a step can cost you money or damage a component, that warning comes before the step, not after it.

We don't recommend a tool we wouldn't use ourselves, and if a link ever earns us anything it will say so on the page.

Who makes it

Lets Build a Bot is produced by Click Shift Marketing, a B2B performance marketing agency in Quebec, alongside sister projects in campaign and demand automation. We build agents for our own work constantly; the projects here are the ones we found worth explaining.

Found a mistake, or built something from a project here? dave@clickshift.ca — corrections get fixed and credited.

Privacy, briefly

If you subscribe, we store your email so we can send you the weekly project, and nothing else. No selling, no other lists, unsubscribe in one click. Ask and we'll delete it.