Hey Reader,

Three reads this week, and two of them are about the same quiet shift: the tool got good enough that the bottleneck moved somewhere less glamorous. An agent that can drive a video timeline still cannot decide what belongs on it. A workflow that looks automatable can still be the wrong one to touch first.

I let Claude Code edit a video, and it never opened a decision

Palmier exposes its macOS timeline as a local MCP server, so the agent reads the real project – tracks, clips, frame positions – and makes the edits itself. On the first full run it took a 47-second passage out of a long recording, placed every clip, split the face footage at each layout boundary, styled the captions onto the seam, then inspected the timeline back to confirm nothing collided. I never opened the editor. What nobody puts in the demo is what that setup is actually like: an agent with timeline access is an extremely capable, extremely literal junior editor who has never seen your work. It places a clip at exactly the frame you name, and it has no idea your top half should never sit empty. So the work moved rather than disappeared, into four layers of specification I had to write first. The honest accounting is in there too – those catalogs took longer than editing the first several clips by hand would have.

Read the whole build →

Five checks that tell you whether a workflow is ready

Score one workflow, not your whole company. How often it happens, how writable the steps are, how much it interrupts, how costly a small mistake is, and how much license or liability weight it carries. Five points each, 25 total. The part worth arguing about is the order: the liability check is evaluated above the total, not alongside it, so a workflow can score 23 out of 25 and still come back as "this stays with a person." Write those two conditions the other way around and you get a tool that recommends automating regulated decisions with flawless arithmetic. One check also reads backwards to almost everyone – higher interruption cost makes a workflow more worth automating, not less, because the real cost is broken focus and a waiting customer, and neither shows up on a time sheet.

Score a real workflow →

All 15 Claude workflows for a small business, with the 2026 numbers

This one is rebuilt for 2026: current pricing, the workflows that actually survived contact with real businesses, and the ones that quietly did not. Fifteen workflows, what each costs to run, and the honest section on when Claude is the wrong tool – which matters more than the list, because most of what gets sold as an AI use case is a spreadsheet with extra steps. If you read an earlier version of this, the numbers and the skip-it section are the parts that changed.

See the 2026 numbers →

Reply and tell me which one was the read for you. I read every reply.

– Alex

P.S. – The conversations behind reads like these happen in the WotAI Skool community. 760+ builders, three live calls a week, free to jump in.

Before you automate the wrong thing

Both of this week's new reads land on the same practical question, and it is not "which tool." It is which piece of work you should hand over first, and that one has an answer you can get to in about five minutes.

Nine questions against one real workflow. It returns a grade, names your weakest check in plain language, and tells you the single number worth measuring before and after. It will also tell you to leave a workflow alone, which is the outcome most tools in this category are structurally incapable of returning.

Pick the process you would fix first and run it through.