When people talk about AI building or updating systems, the conversation often jumps to tools. The quieter truth is older: if your documentation is thin, your handoffs between teams are fuzzy and nobody can measure whether a change worked, adding AI just speeds up the mess.
An engineering leader I spoke with on my podcast put it this way. The same things that serve people — good documentation, clear separation between parts of the system, and solid checks — also let AI build safely and quickly. And those basics matter more now, because competitors who have them are moving at machine speed instead of human speed.
That's the executive takeaway. Foundations first. Speed second.
Foundations are boring until they're missing
In a plant, think of the difference between a changeover with a written procedure and one that lives in one person's head. In logistics, think of exception playbooks versus tribal knowledge on the dock. In E&C, think of drawing packages and RFI history that a new PM can actually find.
Those are human foundations. They're also what AI needs if you want it drafting procedures, suggesting schedule changes or helping update internal tools without inventing its own reality.
The tension: move fast with AI vs. clean the house first
On the one hand, waiting forever to "get ready" is how you fall behind. On the other, pointing AI at undocumented systems and fuzzy ownership is how you get confident wrong answers at scale.
I land in the middle. Don't freeze. Name the three foundations — docs, handoffs, checks — and raise the floor on the workflows you're about to put AI near. Then let speed follow.
What "foundations" means in plain English
- Documentation. How the process works, who owns it, where the source of truth lives.
- Clear handoffs. Where one system or team stops and the next starts, with no overlapping "someone else has it."
- Measurable checks. A simple test that tells you the change worked: uptime, on-time rate, error count, cycle time.
An example
Say an E&C firm wants AI to help draft RFI responses from past projects. If past RFIs live in five folders with no tags, AI will guess. If each project has a clean package — scope, drawings, prior RFIs, who signed what — AI can draft a useful first pass and a PM can approve it in minutes. Same tool. Different foundation.
Or a warehouse that wants AI to suggest exception handling. Without a written playbook and a clear owner per exception type, the suggestions sound smart and still land on the wrong shift lead. With a playbook and checks (mis-sort rate, time-to-resolve), AI can propose and people can trust the review.
Signs your foundations aren't ready
- New hires take months to learn "how we really do it."
- Two teams both think they own the same handoff.
- Nobody can say how you'd know an AI-suggested change worked.
- Fixes depend on one person who isn't on vacation this week.
What to do this week
Pick one workflow you're tempted to put near AI. Score it on three questions: Is the process written down? Is ownership of each handoff clear? Is there a number that proves the change worked? Fix the lowest score before you expand the tool.
Common mistakes
- Buying tools before docs. The tool amplifies whatever you already have.
- Treating this as an IT-only problem. Ops owns the playbooks. IT owns the systems. Both matter.
- Checks nobody looks at. A metric that isn't reviewed is decoration.
- Jargon that scares the room. You don't need to talk about agents in every meeting. Talk about docs, handoffs and checks.
Who should own this
The COO or ops lead owns process documentation and handoffs. The CIO or systems owner owns the technical boundaries and checks. The CEO owns the rule: we don't put AI near a workflow that can't pass the three-question score.
Machine speed needs human foundations
If competitors are moving faster with AI, the gap is often foundations, not model choice. A 15-minute AI roadmap will show which workflows are ready and which need a cleanup first. For the full picture, see our AI readiness assessment.