Writing
Perspective from
the field.
I write about enterprise AI, technology leadership, and the uncomfortable truths most people in this industry won't say out loud. No vendor pitches. No hype cycles. Just what I actually see.
What I cover.
My writing sits at the intersection of enterprise technology, organizational leadership, and honest field observation. I have spent 25 years building and breaking large-scale systems, plus a term as Mayor of a real city. When I write, I am drawing on both.
Enterprise Field Notes.
Published weekly as a newsletter on LinkedIn, with the full version here on benpickett.com. Click any card to read the full piece.
Enterprise Field Notes
Weekly dispatches on enterprise AI strategy: what's working in the field, what's failing, and what it means. No vendor pitches. Published on LinkedIn, with the full version on benpickett.com.
The Enterprise Field Notes series
Oracle Sped Up Coding With AI. The Wait Moved to Release. Has Yours?
Oracle reached 80 percent adoption of its AI coding tools in three months, and its co-CEO told staff that quicker code hadn't made everything faster. Where the wait moved, the test environments I rebuilt years ago, and five questions to find yours.
Databricks Piloted OpenAI’s New Model With About 200 Engineers. Their Coding Spend Rose About 60 Percent, and It Got Its Own Sub-Budget.
Databricks measured what a new model did to spend on about 200 engineers before 3,500 had it. The same month, four vendors changed what AI usage costs. And the contract I once found that nobody had checked against what we were actually doing.
Walmart, Home Depot and KPMG Already Had What They Needed. So Do You.
I went back through twenty issues and found one story told nine times. Walmart, Klarna, KPMG, Ford, Amazon, Home Depot, Lowe’s, Meta and OpenAI all had what they needed. What was missing was somebody told to go and look.
OpenAI’s Agents Got Caught the Way Teenagers Get Caught
They wiped the whole thing, and about a day later there was a new one built out of directory names. What that breaks about the word resolved, the two numbers OpenAI published that nobody else has, and the signatory count I could not hold still long enough to print.
Meta Changed Its Internal Tools 220% More. Major Incidents Rose 40%.
Meta put generative capacity into its engineering org and wrote down both halves of what happened. Code changes to internal tools up 220 percent, delivered features up 36, major incidents up 40, remediation time up 70. Why the rate got better and the bill got worse, and the number your company is not putting next to its own.
Target Hired Lowe's AI Chief. His Page Was Gone in Six Days.
Target hired the man who built Lowe's AI. His page was gone from the Lowe's leadership site six days later, and nobody said his name on the earnings call. What a log records, what it cannot, and the habit that has no line item.
Anthropic's Own Safety Test Broke Into Three Companies. Nobody Noticed, Not Even Anthropic.
Anthropic searched 141,006 of its own test runs to find three companies its AI had broken into. The companies did not know. Neither did Anthropic. A chemical plant, a missing instrument, and the afternoon that tells you where you stand.
Home Depot Had AI in 600 Stores Six Weeks After ChatGPT Launched. It Was Not Watching ChatGPT.
I led the teams that built systems people could not use, and we built them from a conference room. A randomized trial, Census Bureau data on American manufacturers, and five things I got wrong at scale.
Amazon Slowed Down on Purpose. A Founder Lost Three Months in Nine Seconds.
You already built the rule that would have stopped it. Nobody ever decided whether it covers the agents you switched on this year, and that decision is an onboarding problem, not an AI safety problem.
OpenAI's Own AI Broke Into Hugging Face. The AI Its Defenders Called for Help Refused.
OpenAI's own AI escaped its sandbox and broke into Hugging Face over a single weekend. When the defenders called for help, the AI they turned to refused. Tools used to fail one way, they stopped. AI can now act on its own and it can refuse you, and the controls we trust were built for neither.
Two AI Models Went Dark Worldwide in June. Yours Could Be Next.
For nearly three weeks, one government order took two frontier AI models offline for every company that ran them. The model your business runs on is no longer just a technology choice, it is a political one that someone else can switch off. Here are four moves that keep you running when the next one goes dark.
Anthropic Just Showed You Your AI Has No Backup
Anthropic pulled two of its own frontier models overnight, and most companies running on them had nothing ready to switch to. The lesson is not about one vendor. It is that the backup you make before you need it is the only one that ever counts.
Ford Rehired the People AI Was Supposed to Replace
A year ago Ford's CEO said AI would replace half of all white-collar workers. This year it hired back 350 veteran engineers, because its AI was not good enough. AI did not fail Ford. Using it to replace hard-won expertise did.
KPMG Forgot to Check Its Own Work
KPMG, one of the most trusted firms on earth, shipped a report with 45 sources and only five checked out. It is not really about KPMG, or the model. It is the one human check that quietly stops firing the moment AI makes a draft look finished.
The SpaceX IPO Is a Bet on Your AI Bill
Intelligence keeps getting cheaper, yet most teams' AI bills keep going up. The fix is not a cheaper model, it is knowing when to stop reaching for the biggest one. The same move that pulls your bill down also eases what AI takes from the grid and the river.
What Klarna Got Wrong That Walmart Got Right
Klarna said its AI did the work of 700 agents, then quietly started hiring people back. Walmart took the opposite path and pulled ahead. The difference was not the technology, it was building AI around people instead of in place of them.
What Walmart Understood That the Other 95% Didn't
Walmart published the blueprint for enterprise AI, and almost nobody outside retail has read it. Four super agents, an orchestration layer, and the part you can copy without their two years and five thousand engineers.
Dear Andrew, COO of Uber
An open letter to Uber's COO, who said on the record that he cannot draw a line between what Uber spends on AI and what it produces. Why that gap is an architecture problem, not a measurement one.
The Definitions Gap
Teams do not fail at AI because the model is wrong. They fail because they never agreed on what the words mean. The shared-language problem sitting upstream of every AI initiative.
The Productivity Disconnect
Workers say AI changed nothing. Executives project massive gains. Both are telling the truth. The gap is a multiplication problem, and it is where the productivity actually went.
The Single Model Trap
One default model, every task, and the apology loop that follows when it is wrong. Why routing each task to the right model is the highest-leverage decision in enterprise AI right now.
The Pilot to Production Gap
MIT reviewed 300+ AI initiatives and found only 5% reach production at scale. What separates the 5% from everyone else, plus the eight-question audit to run before your next AI evaluation.
The AI Strategy That Was Actually Running
78% of the people already using AI at work are on tools IT never sanctioned. The real AI strategy in most companies is the one employees adopted without anyone approving it. Blocking it makes everything worse.
More articles
Anthropic's Mythos Just Proved Enterprise AI Governance Is a Cybersecurity Emergency
An AI escaped containment, found thousands of zero-day vulnerabilities autonomously, and triggered a Fed-level response. Machine speed doesn't wait for human response time. The gap between them is your problem.
Read here →Anthropic's Two Leaks in Five Days: What Enterprise Security Leaders Must Do Now
The company warning the U.S. government about AI cybersecurity risks shipped a misconfigured CMS and an unstripped Claude Code source map within the same week. A case study in the gap between AI security posture and AI security operations.
Read here →AI Is an Accelerator, Not a Replacement for People
Companies augmenting human capabilities outperform those eliminating roles, and 50% of organizations making AI-driven cuts are expected to rehire by 2027. The workforce reduction narrative around AI is not just ethically fraught. It is bad strategy.
Read here →