https://loyokelley.com/insights en-us Copyright 2026 Loy O'Kelley Field notes on running enterprise programs with AI in the loop — governance, reporting, program risk, and where the machine stops and judgment starts. Narrated editions of the essays at loyokelley.com. Loy O'Kelley Field notes on AI-augmented program delivery from a senior transformation delivery leader. episodic false Loy O'Kelleyloy@loyokelley.com https://loyokelley.com/the-plan-nobody-agreed-to https://loyokelley.com/the-plan-nobody-agreed-to Wed, 09 Sep 2026 13:00:00 +0000 AI drafted the program plan in ten minutes and it was better than the one the team would have built by hand. Six weeks later, three of the owners on it admitted they'd never really looked at their line. The schedule was never the deliverable — planning's actual product was a room full of people arguing until each of them had committed to something out loud. AI drafted the program plan in ten minutes and it was better than the one the team would have built by hand. Six weeks later, three of the owners on it admitted they'd never really looked at their line. The schedule was never the deliverable — planning's actual product was a room full of people arguing until each of them had committed to something out loud. Loy O'Kelley 2:28 false https://loyokelley.com/common-operating-picture https://loyokelley.com/common-operating-picture Wed, 02 Sep 2026 13:00:00 +0000 Your CFO asked a copilot how the program is doing. So did your sponsor, your PMO lead, and your integrator's account exec — different question, different context, different answer. The status meeting used to be where everyone received the same picture. Now everyone arrives pre-briefed, and no two briefings match. Your CFO asked a copilot how the program is doing. So did your sponsor, your PMO lead, and your integrator's account exec — different question, different context, different answer. The status meeting used to be where everyone received the same picture. Now everyone arrives pre-briefed, and no two briefings match. Loy O'Kelley 6:29 false https://loyokelley.com/the-machine-that-cried-wolf https://loyokelley.com/the-machine-that-cried-wolf Wed, 26 Aug 2026 13:00:00 +0000 You wired AI into the program to warn you sooner, and it worked — it warns you about everything, constantly. Forty flags a week later, your team has learned the survival skill that kills early-warning systems: scrolling past. An alert nobody acts on isn't a control. It's noise with a timestamp. You wired AI into the program to warn you sooner, and it worked — it warns you about everything, constantly. Forty flags a week later, your team has learned the survival skill that kills early-warning systems: scrolling past. An alert nobody acts on isn't a control. It's noise with a timestamp. Loy O'Kelley 6:36 false https://loyokelley.com/recorder-in-the-room https://loyokelley.com/recorder-in-the-room Wed, 19 Aug 2026 13:00:00 +0000 Somewhere in the last eighteen months, every meeting on your program quietly became a verbatim, searchable, permanently retrievable record. Nobody proposed that. No governance body approved it. It arrived one bot at a time, and it changed both what gets said in the room and what your organization can be held to afterward. Somewhere in the last eighteen months, every meeting on your program quietly became a verbatim, searchable, permanently retrievable record. Nobody proposed that. No governance body approved it. It arrived one bot at a time, and it changed both what gets said in the room and what your organization can be held to afterward. Loy O'Kelley 7:35 false https://loyokelley.com/subcontracted-judgment https://loyokelley.com/subcontracted-judgment Wed, 12 Aug 2026 13:00:00 +0000 Enterprises have spent two years building AI policy for their own people — approved tools, usage standards, review gates. Then they buy sixty percent of the program from a system integrator whose delivery model changed under them, quietly, sometime last year. Your governance stops at your badge line. Your name on the outcome does not. Enterprises have spent two years building AI policy for their own people — approved tools, usage standards, review gates. Then they buy sixty percent of the program from a system integrator whose delivery model changed under them, quietly, sometime last year. Your governance stops at your badge line. Your name on the outcome does not. Loy O'Kelley 6:57 false https://loyokelley.com/chain-of-custody https://loyokelley.com/chain-of-custody Wed, 05 Aug 2026 13:00:00 +0000 Six months after a number lands in a steering deck, somebody asks where it came from. When a human built it, the trail existed — a file, a query, a person who remembered the assumption. When an agent built it, there's often nothing behind the number but the number. That gap isn't a documentation problem. It's the difference between a program that can defend itself and one that can only apologize. Six months after a number lands in a steering deck, somebody asks where it came from. When a human built it, the trail existed — a file, a query, a person who remembered the assumption. When an agent built it, there's often nothing behind the number but the number. That gap isn't a documentation problem. It's the difference between a program that can defend itself and one that can only apologize. Loy O'Kelley 7:26 false https://loyokelley.com/judgment-atrophy https://loyokelley.com/judgment-atrophy Wed, 29 Jul 2026 13:00:00 +0000 Nobody learned to run a program by running one. You learned it by building a hundred schedules that were wrong, reconciling numbers that refused to reconcile, and sitting through the meeting where your estimate fell apart. AI is now doing all of that — and doing it well. The work goes away. So do the reps that used to turn a coordinator into someone you'd trust with a $40M recovery. Nobody learned to run a program by running one. You learned it by building a hundred schedules that were wrong, reconciling numbers that refused to reconcile, and sitting through the meeting where your estimate fell apart. AI is now doing all of that — and doing it well. The work goes away. So do the reps that used to turn a coordinator into someone you'd trust with a $40M recovery. Loy O'Kelley 6:59 false https://loyokelley.com/the-confidence-problem https://loyokelley.com/the-confidence-problem Wed, 22 Jul 2026 13:00:00 +0000 A human who isn't sure sounds like it — the pause, the "I think," the caveat before the number. That hedge is information, and experienced leaders run on it. AI-drafted reporting deletes the hedge: it sounds exactly as certain reading solid data as it does papering over a gap. If you don't engineer the doubt back in, your program loses one of its oldest early-warning systems. A human who isn't sure sounds like it — the pause, the "I think," the caveat before the number. That hedge is information, and experienced leaders run on it. AI-drafted reporting deletes the hedge: it sounds exactly as certain reading solid data as it does papering over a gap. If you don't engineer the doubt back in, your program loses one of its oldest early-warning systems. Loy O'Kelley 6:38 false https://loyokelley.com/the-seams https://loyokelley.com/the-seams Wed, 15 Jul 2026 13:00:00 +0000 AI made the inside of every team faster and easier to see — the board, the burndown, the automated status. But programs don't break inside teams. They break at the seams between them: the handoff nobody owns, the dependency with two owners and therefore none, the interface that drifted while both sides looked away. You can automate all the work and still lose the program at the joint. AI made the inside of every team faster and easier to see — the board, the burndown, the automated status. But programs don't break inside teams. They break at the seams between them: the handoff nobody owns, the dependency with two owners and therefore none, the interface that drifted while both sides looked away. You can automate all the work and still lose the program at the joint. Loy O'Kelley 7:22 false https://loyokelley.com/automation-debt https://loyokelley.com/automation-debt Wed, 08 Jul 2026 13:00:00 +0000 Standing up an automation feels like deleting work. It isn't. Every workflow you wire up is a new thing that can break silently, drift out of date, and fail at the worst possible moment — and someone's name has to be on keeping it alive. The build is the cheap part. The upkeep is the program. Standing up an automation feels like deleting work. It isn't. Every workflow you wire up is a new thing that can break silently, drift out of date, and fail at the worst possible moment — and someone's name has to be on keeping it alive. The build is the cheap part. The upkeep is the program. Loy O'Kelley 6:51 false https://loyokelley.com/pilot-purgatory https://loyokelley.com/pilot-purgatory Wed, 01 Jul 2026 13:00:00 +0000 Your AI pilot worked. Congratulations — you've cleared the easy 10%. A demo runs on curated data, one willing user, and no consequences. Production runs on the real board, a skeptic, an integration nobody scoped, and someone's name on the output. That gap is where most AI programs quietly die. Your AI pilot worked. Congratulations — you've cleared the easy 10%. A demo runs on curated data, one willing user, and no consequences. Production runs on the real board, a skeptic, an integration nobody scoped, and someone's name on the output. That gap is where most AI programs quietly die. Loy O'Kelley 6:15 false https://loyokelley.com/rubber-stamp-review https://loyokelley.com/rubber-stamp-review Wed, 24 Jun 2026 13:00:00 +0000 Keep a human in the loop and you can tell yourself the work is supervised. But the better the model gets, the less the reviewer actually reads — and a checkpoint that has quietly stopped looking is worse than none, because the whole organization still trusts it. Keep a human in the loop and you can tell yourself the work is supervised. But the better the model gets, the less the reviewer actually reads — and a checkpoint that has quietly stopped looking is worse than none, because the whole organization still trusts it. Loy O'Kelley 7:14 false https://loyokelley.com/decision-backlog https://loyokelley.com/decision-backlog Wed, 17 Jun 2026 13:00:00 +0000 A program rarely dies because the work was too hard. It dies in the gap between when a decision was needed and when someone finally made it. Your task board tracks the work down to the hour. Almost nobody tracks the decisions — which is exactly where the schedule quietly bleeds out. A program rarely dies because the work was too hard. It dies in the gap between when a decision was needed and when someone finally made it. Your task board tracks the work down to the hour. Almost nobody tracks the decisions — which is exactly where the schedule quietly bleeds out. Loy O'Kelley 6:37 false https://loyokelley.com/shadow-ai-on-your-program https://loyokelley.com/shadow-ai-on-your-program Sun, 14 Jun 2026 13:00:00 +0000 You don't get to decide whether AI is on your program. Your team already made that call — quietly, in a browser tab, pasting a status update into a chatbot to make the Friday deadline. The only thing leadership still controls is whether that use is governed or invisible. A ban doesn't remove the risk. It removes your ability to see it. You don't get to decide whether AI is on your program. Your team already made that call — quietly, in a browser tab, pasting a status update into a chatbot to make the Friday deadline. The only thing leadership still controls is whether that use is governed or invisible. A ban doesn't remove the risk. It removes your ability to see it. Loy O'Kelley 7:41 false https://loyokelley.com/ai-accountability https://loyokelley.com/ai-accountability Wed, 10 Jun 2026 13:00:00 +0000 AI now drafts the report, flags the risk, and moves the ticket. None of that changes who owns the outcome. The accountability boundary is the one design decision that determines whether AI makes your program safer — or just faster on the way to the cliff. AI now drafts the report, flags the risk, and moves the ticket. None of that changes who owns the outcome. The accountability boundary is the one design decision that determines whether AI makes your program safer — or just faster on the way to the cliff. Loy O'Kelley 5:20 false https://loyokelley.com/garbage-in-gospel-out https://loyokelley.com/garbage-in-gospel-out Wed, 10 Jun 2026 13:00:00 +0000 Point an agent at a dishonest Jira board and it won't tell you the board is lying. It will summarize the lie — fluently, confidently, in executive-ready prose. Before AI can make your program smarter, your system of record has to deserve the trust you're about to automate. Point an agent at a dishonest Jira board and it won't tell you the board is lying. It will summarize the lie — fluently, confidently, in executive-ready prose. Before AI can make your program smarter, your system of record has to deserve the trust you're about to automate. Loy O'Kelley 6:21 false https://loyokelley.com/ai-augmented-pmo https://loyokelley.com/ai-augmented-pmo Sun, 31 May 2026 13:00:00 +0000 Gartner says AI could absorb up to 80% of traditional project management tasks by 2030. The number is real. The conclusion most people draw from it is wrong. Gartner says AI could absorb up to 80% of traditional project management tasks by 2030. The number is real. The conclusion most people draw from it is wrong. Loy O'Kelley 5:34 false https://loyokelley.com/ai-delivery-stack https://loyokelley.com/ai-delivery-stack Sun, 31 May 2026 13:00:00 +0000 Not a tool list. A working philosophy: the model thinks, the automation moves, and a human owns the boundary between them. Here's what I actually run, and why. Not a tool list. A working philosophy: the model thinks, the automation moves, and a human owns the boundary between them. Here's what I actually run, and why. Loy O'Kelley 5:42 false https://loyokelley.com/ai-program-risk https://loyokelley.com/ai-program-risk Sun, 31 May 2026 13:00:00 +0000 The expensive thing about a program collapse is rarely the collapse. It's the three months everyone spent not admitting it. That's the window AI actually changes. The expensive thing about a program collapse is rarely the collapse. It's the three months everyone spent not admitting it. That's the window AI actually changes. Loy O'Kelley 4:41 false https://loyokelley.com/status-report-tax https://loyokelley.com/status-report-tax Sun, 31 May 2026 13:00:00 +0000 A program doesn't fail at the steering committee. It fails in the two weeks of reality the steering committee never heard about. AI closes that gap — if you build it to tell the truth, not just to be fast. A program doesn't fail at the steering committee. It fails in the two weeks of reality the steering committee never heard about. AI closes that gap — if you build it to tell the truth, not just to be fast. Loy O'Kelley 4:37 false