What's production-ready in AI automation right now
A field guide to what's solid, what's hype, and what quietly breaks in week two, once the demo's over.
Every year the automation toolkit gets better and the marketing around it gets worse. Here's what's worth reaching for on a real client project this quarter, and what's still worth steering around.
Solid: retrieval over your own docs
RAG stopped being a research topic a while ago. With a decent chunking strategy and a reranker, answering support or onboarding questions from your own knowledge base is boring, reliable engineering now, which is exactly what you want from something customers touch.
Solid: narrow agents with a short tool list
An agent with three or four well-scoped tools and a clear stop condition is easy to reason about and easy to monitor. The failure mode isn't the model. It's teams giving an agent twelve tools and hoping.
Still shaky: fully autonomous multi-step workflows with no review gate
Chaining five agent calls together with no human checkpoint looks great in a demo video. In production, the error compounds at every step. If a workflow makes a decision a client would want to see before it ships, put a review gate there. That's not a limitation of the tooling. It's just good systems design.
Hype: “fully autonomous” anything, sold to a non-technical buyer
If a pitch deck says “set it and forget it,” ask what happens when the input data looks nothing like the demo data. If there's no good answer, that's the tell.
Get new posts by email
AI updates, trend breakdowns, and picks for the certifications worth your time, a few times a month, when something's worth sending.
Have something like this to automate?