From William
Hi — I’m William McBride. I spent 18 years building technology for large enterprises; now my mission is to democratize those lessons for small and medium-sized businesses — because owners deserve the same operating discipline the Fortune 500 gets.
Most AI conversations start with the same question: what can the system do for us?
I think the better first question is simpler: what decision are we trying to make easier?
One recent conversation and a public-stage session back at CES in January pointed to the same answer from very different places. The first was about a large organization trying to prove the history behind a time-sensitive procurement request. The CES session was about the order a major enterprise chose when it put AI into real operations.
Different scale. Same operating problem. The work was not really about AI replacing judgment. It was about making context easier to find, commitments easier to confirm, and decisions easier to trust.
The story: AI’s first job is recall
The team had raised an infrastructure need months earlier. By the time the issue escalated, the commercial environment had changed and the new quote was materially higher.
The deciding question was not technical. It was operational: could the team prove when the need had been raised, what had been requested, and why the timeline mattered?
AI-assisted search and summarization helped reconstruct the relevant record trail. People still checked the facts. People still made the decision. People still owned the outcome.
The AI did not negotiate the contract. It did not approve the purchase. It did not send a message on anyone’s behalf. It made the organization’s memory easier to use at the moment the decision needed evidence.
The outcome was concrete. With the documented history in hand, the organization kept its earlier pricing instead of accepting a re-quote — a six-figure difference. Timing would have cut the other way: hardware prices had climbed since the original request, and a fresh quote would have carried every one of those increases.
That is why this is more than a records story. Managers everywhere are watching hardware costs rise and seeing those increases land directly in quotes, renewals, and project budgets. In that environment, being able to prove what you asked for — and when you asked — is not administrative hygiene. It is negotiating leverage.
Three takeaways worth keeping
- Recall beats automation as the first win. The first valuable AI workflow is often not an agent — it is a better memory layer around your own records. Capability is getting cheap this year; context about your business is not — and context is the part only you can supply.
- The rule is not the control — the connection is. Telling a tool “never send messages” is a sentence in a prompt; connecting it to an account that cannot send is a control. Start read-only, on records you control — the open inbox and open web can come later.
- “Should we?” beats “could we?” Every vendor can say what a system could do; only your operation can say what it should do. Start from a decision that is expensive when context is missing, not from a demo that looks impressive.
Field note · From a CES enterprise-adoption panel
Search before agents
One large operator’s real adoption sequence was quieter — and more useful — than the demos.
Back in January at CES, an enterprise IT leader walked through the adoption path her organization had taken since 2023. The first major step was not an autonomous agent. It was enterprise search for call-center representatives.
That changed how representatives supported customers because answers no longer depended as heavily on who happened to remember where the information lived. The organization then introduced general-purpose copilots and self-service tools. Agents came later, paired with a clear list of prerequisites: infrastructure guardrails, identity, security, and performance.
Seven months later, it is the advice from that week that has aged the best. The flashy demos have mostly churned. The sequence has held.
Strip away the enterprise vocabulary and the sequence is owner-sized. Search first means the quote you texted a customer in March is findable before you buy anything that drafts replies — and when a customer disputes a price, you can show you quoted $4,200 before material costs moved. That is the procurement team’s leverage, at the scale of one truck and one phone.
A small company can start by making customer histories, commitments, quotes, and approvals searchable. It can then use AI to summarize or draft while a person reviews the result. Only after that workflow is reliable should the system receive permission to act.
That is not slower adoption. It is adoption in an order that gives each new capability something trustworthy to build on.
The other voice on the same panel. A second executive on that panel runs a retail network of store leaders and franchisees. Shortly before CES, her team brought her a proposal, and she stopped it with one question — that is not even a problem we have, so why would we deploy the technology? Early on, she said, the discipline is not what could we do. It is what should we do.
She added two tests an owner can use unchanged. Her team’s tolerance for clunky technology is zero, so anything new has to work like a silent partner, not one more tool to jump between. And some interactions get marked human-to-human before any automation starts: someone hitting the panic button in a store does not want a bot. A franchisee running one location is a small business — this is not enterprise advice scaled down.
The risk is not that AI cannot be useful. The risk is that teams skip straight to automation before they know what the system is allowed to remember, who is allowed to see it, and who owns the final action. No tool owns an outcome. Someone still signs — and that someone is you.
The 20-minute move
Pick one active customer request, vendor decision, appointment, or internal project where missing context would create real cost. Not the tidy one — the expensive one. Then ask:
- Where is the record? One reliable source, or inboxes, texts, notes, and memory?
- Who can access it? Is the access model intentional, or just inherited?
- Who reviews the output? If AI summarizes the history, who checks it before anyone acts?
- What action is allowed? Suggest, draft, and retrieve only — or change records and send messages?
If those answers are unclear, that is the first use case — the current workflow is asking people to operate from memory.
What I’m watching
The organizations that benefit from AI first will not necessarily have the most aggressive automation roadmap. They will make their information usable, their commitments confirmable, and their approvals intentional.
That is the standard I am building toward with ARIA: help people prepare, retrieve, organize, and follow up while keeping consequential actions under the user’s control.
Reader question
What commitment in your business would be cheaper or safer if you could recover the full context behind it in five minutes?
It might be a customer request, a vendor promise, or an internal approval. The answer is probably a better first AI use case than another generic automation demo.