KM

First Build the Machine. Then Add the AI Engine.

Published 4 min read
  • AI
  • AI tooling

Most companies add AI before they have a workflow worth amplifying. AI is not the system. It is the amplifier. Structured processes create scalable AI. Broken ones create faster chaos.

First Build the Machine. Then Add the AI Engine.

Most companies are trying to add AI before they build a workflow worth automating.

That is backwards.

AI is not the system. AI is the amplifier.

If your business runs like a tinkering lab (scattered tools, unclear handoffs, inconsistent process), AI will not fix it. It will scale the mess.

I have been arguing adjacent points in AI should enhance the system, not subsidize unclear thinking and a better model will not fix unclear thinking. This post is the operating sequence: machine first, engine second.

What “amplifier” looks like in practice

In the figure above, left is the day most teams actually live: sticky notes instead of process, a desk of exceptions, and a constellation of apps (Sheets, Slack, Notion, Trello, ChatGPT) connected to the same red tangle. Notice ChatGPT is already in the chaos. Buying another model does not untangle that picture. It adds another dashed line into the scribble.

On the right, the same person after the sequence is reversed: a defined path, a human checkpoint where judgment belongs, measurable completion, and AI sitting on top of that path as power, not as a substitute for knowing what the work is.

That is the difference between amplifying a machine and amplifying a mess.

The research keeps saying the same unsexy thing

McKinsey’s State of AI survey found that among the organizational attributes tested, workflow redesign had the biggest effect on an organization’s ability to see EBIT impact from gen AI. Only about 21% of respondents whose organizations use gen AI said they had fundamentally redesigned at least some workflows.

Most firms adopted the tools. Fewer rebuilt the work. The financial signal follows the rebuild, not the logo on the invoice.

McKinsey’s older automation guidance is just as blunt: treating automation as a technology-led effort can doom a program. Process problems are rarely fixed by dropping in a new technical solution. Poor input quality, too many variations, and “off script” procedures tend to undermine the project once implementation starts (How to avoid the three common execution pitfalls that derail automation programs).

Their agentic work makes the same point in newer language: embed agents into a legacy process without redesign and you usually get a faster assistant inside a still-sequential bottleneck (Seizing the agentic AI advantage).

Amplifiers do what amplifiers do. They raise the volume of whatever you feed them.

The real sequence

  1. Build the workflow. Name the steps, owners, inputs, outputs, and exceptions. If two people cannot draw the same diagram, you do not have a workflow. You have folklore.
  2. Remove friction. Cut steps that exist because of history, ego, or tool sprawl. Fix handoffs. Decide which exceptions are real and which are laziness dressed as edge cases.
  3. Then apply AI. Put the model where the work is repetitive, high-volume, or language-heavy, inside boundaries you already understand.

Structured systems create scalable AI. Broken systems create faster chaos.

Where AI implementation actually works

AI implementation works best where three conditions already exist:

  • Process is defined. Not in a dusty SOP nobody opens. In the way work actually moves this week.
  • Approvals are clear. Who can say yes, who can say no, and what must escalate.
  • Outcomes are measurable. Cycle time, error rate, first-pass yield, cost per unit. If you cannot baseline it by hand, you cannot tell whether the engine helped.

Before AI power comes workflow power.

If you cannot run the path manually for two clean cycles and write it on one page, you are not ready for automation. You are ready for process work. That sentence will offend people who want a tool launch to replace operational discipline. It should.

First build the machine. Then add the engine.

An engine on a pile of parts is not a vehicle. It is a fire hazard with a warranty.

So before the next “AI initiative,” ask the only diagnostic that matters: do we have a machine worth powering, or are we hoping the engine will invent the chassis?

If the honest answer is the second, stop shopping for models. Start mapping the work.

References