I'll be at Make Waves '26 to talk all things AI and automation workflow strategy. If you're attending or in town, let's meet up at the Playground or the afterparty! 🎉
Link: https://t.co/U79EKy5nRs #MakeWaves26#AI#Automation
Three non-negotiable standards every ops director should implement before scaling beyond 50 team members: 1. Clear single-owner accountability for each workflow. 2. Real-time alerting on SLA breaches. 3. Zero manual data re-entry across tools.
No one who truly wants to build a reliable AI backed system should be using OpenAI models. From being moral police on stupid instances to repeatedly setting its own override decisions, this is a total nightmare and a bad product anyone should be paying for.
Stripe already handles card updates and invoices. Turning basic billing into support tickets is a product choice, and it makes taking a week off harder than it should be.
Before I automate a workflow, I want to watch it run by hand. The spreadsheet nobody documented and the Friday screenshot often tell me more than the process document.
A skill is a prompt after production has beaten the pretty theory out of it. I do not save one because it sounds clever in a blank doc. I save it after live work, audit, rewrite, and another run against actual inputs.
Founders want software margins.
Customers want the thing to work.
In messy markets, the moat may be the work everyone calls “not scalable”: training users, fixing edge cases, rewriting the workflow, and sitting with the customer until the AI is actually useful.
Memory can make an agent useful, but unmanaged memory turns into compost. The maintenance job needs rules for deleting stale notes, tightening skills, checking old assumptions, and keeping approvals visible.
Pay per crawl is a serious hint of a web where the request itself can carry money. A page, software access point, data file, or tool call can price access before the request reaches the origin server.
The AI service companies I bet on look like ordinary agencies; Interior design, bookkeeping, recruitment, and legal support.
The AI sits quietly behind the work and reduces the number of hours needed to deliver results.
The phrase "AI touched this" is too vague for policy. Spellcheck, image generation, voice cloning, a chatbot, and an AI summary do not carry the same risk.
Provider and deployer are separate jobs. OpenAI can add metadata to a file, but the business publishing a cloned voice still owns the visible disclosure problem.