Unpopular opinion: AI agents are going to replace most of what we are learning to build here

Asked by Automate Naija in general · 2 replies

Saying this out loud in an automation cohort feels a little like setting fire to the building, but here it is.

A hand-wired n8n workflow is a set of decisions someone worked out in advance: if this, then that, forever, until you go back and edit it. An AI agent that can read an API doc, call the right endpoint, and decide what to do next does not need that map drawn out ahead of time. Give it a goal and some tools, and it improvises the workflow. A lot of what we spend hours wiring node by node, an agent can arguably do by just being told what the outcome should be.

Here is the honest counter to my own point. Agents are non-deterministic. For a payroll run, an invoice, or anything where "it worked differently this time" is not an acceptable answer, you want the boring, predictable, same-every-time workflow, not a model improvising. And somebody still has to build and maintain the tools and workflows an agent calls when it does decide to act. The skill does not disappear, it moves up a layer, from building the whole flow to building the reliable pieces an agent reaches for.

Agree, disagree, or somewhere in between? If you think I am wrong, tell me exactly where.

Replies

Reply by lindaokike

I’m somewhere in the middle. I think agents can take a lot of the manual wiring out of workflows, but for anything where reliability really matters, I’d still trust a deterministic workflow.

Reply by Zoom Play

Seeing what agents can do so far, I think this post isn’t far from thr truth.