Happy Friday friends, and an overdue one. I’ve spent the last couple of weeks chasing the tail end of summer, and then university started back up on me. This semester I’m teaching three classes of advanced Python, plus Digital Solutioning, while building three brand new post-secondary courses on Intelligent Process Automation from scratch. Sorry it’s been so quiet!
I did have time to join an HBR webinar and one slide stopped me cold.
Before we get to that, let me take you back to 2013, and the sharpest room I have ever stood in.
I was part of the Enterprise Markets Team for Apple. I sat across the table from nine executives of a large Canadian airline. The CEO, his CFO, his CIO, his head of people, the president of a new regional carrier, and the person tasked with building a discount airline from scratch. You don’t send nine executives to Cupertino for a chat. You send them because you think you’re standing at the edge of something.
They had a taste of early traction on their consumer app, just enough to spark the fire. The CEO wanted to talk about what was next.
The smartest thing said all day didn’t come from the CEO. It didn’t come from me. It came from one of his EVPs, brand new to the team. When half the room reached for innovation and disruption and a phrase just starting to get fashionable, “digital transformation,” she reached for the opposite.
“For years all we heard was grow, grow, grow. Now all we hear is process, process, process.”
That was the whole thing, right there, in 2013. These were not foolish people. For years, “add more” had meant exactly that: one more process, one more system, one more policy, each one reasonable on its own, each one restricting growth and capabilities. That didn’t stop them from bringing the same premise to Apple’s boardroom. That’s the trap: the smarter the room, the easier it is to mistake “add more” for progress.
Let’s break it down.
Signal:
SHE IDENTIFIED IT BEFORE IT HAD A NAME.
On August 27, I watched an HBR webinar called How Agentic AI is Reshaping Competition, with Vivian Lee, an executive fellow at Harvard Business School and former CEO of University of Utah Health, and Linda Mantia, a board director at Alan, Liberty Mutual, Prove Identity, and Saputo, and Manulife’s former COO. Partway through, they put a name to something I’ve been circling since that boardroom in Cupertino.
Organizational plaque. Their term, not mine, and it’s exact. Established companies build layer upon layer of legacy technology: new tech bolted onto old tech, siloed data, fragmented workflows, segmented decision making. Together, plaque. It narrows the artery until nothing moves through it cleanly, and most leaders don’t notice it’s there because it built up one reasonable decision at a time.
That airline EVP wasn’t talking about AI in 2013. She was describing plaque before anyone had scanned for it. Grow, grow, grow builds the arteries. Process, process, process is what clogs them, one well-intentioned policy, one new system, one workaround at a time, until the org can barely move.
Now AI is the new layer everyone wants to add. And Mantia named the actual decision leaders are facing:
“One of the toughest questions facing leadership is whether to bolt on a new capability or to take on the very tough task of reimagining it and rewiring a company.”
Bolting on is the easy path. You wrap AI around your existing process, you get some productivity lift, you keep the plaque exactly where it was. It works, for a while. It looks safe, until a competitor without your arterial buildup delivers a materially better experience at a materially different cost, and your safe choice turns out to have been the risk the whole time.
Rewiring means starting from the outcome you want and working backward, not from the org chart you already have. It’s the harder path. It’s also the only one that actually removes plaque instead of building on top of it.
Scale:
THE 4 WAYS, AND THE ONE EVERYONE GRABS FIRST.
The webinar laid out four categories most organizations use when they put large language models (LLMs) to work.
Managing information. Digesting and synthesizing complex data. Think insurance prior authorization: parsing hundreds of pages of medical records against policy criteria to find the gaps.
Generating content. Producing text, voice, video. Interactive avatars handling pieces of customer communication.
Translating. Not just languages and dialects, but converting technical, legal, or medical jargon into plain language, and increasingly, natural language into code.
Automating and acting. Delegating repetitive, tedious tasks to autonomous agents.
Ask most leaders which of the four they’re chasing, and it’s the fourth one, nearly every time. Automating and acting is the one that looks like plaque removal. It’s visible, it’s dramatic, it makes a good slide. Meanwhile managing information, the least glamorous of the four, is quietly doing more of the actual unclogging. It’s the one that forces you to confront what your data actually says, which is usually the first honest look your organization has taken at itself in years.
Here’s the part I actually want to talk to you about, though. Not which of the four you pick. How you pick.
In my experience, leaders land on a candidate process for one of two reasons, and neither of them is a good one. Either it’s cheap and easy, the low-hanging fruit that requires no real evaluation. Or someone on the team already ran an unapproved pilot, an executive caught wind of it, and now it gets blessed after the fact because it’s already halfway built. Shadow IT gets a promotion.
Both paths skip the same step. Governance and judgment never get asked to the table. Nobody evaluated whether this was the right process to touch first. It just happened to be the one that was easiest, or the one that got caught.
The webinar’s answer to this is worth stealing outright. Good candidates are high friction and tedious, the kind employees will welcome relief from. They’re places where accuracy matters, because AI tends to outperform tired humans on exactly that. They’re cheap to attempt. And they get audited across three dimensions before anyone touches them: data usability, scale versus complexity, and customer base vulnerability.
Mantia put the discipline in one line I’ve been repeating to my own students since I heard it:
“The goal is not to move fast everywhere. Let’s move fast where we can, carefully where we must, and, importantly, increase our capacity for change.”
That’s the mindful part. Not moving slowly. Moving deliberately, on the process that earns it, instead of the process that happened to be lying around.
Deep Dive:
SOLVE PROBLEM ONE.
The webinar’s case study is NTT Data, and it’s a clean example of doing this right.
NTT competes for large, complex contracts, and they chose to focus their first agentic AI use case on responding to RFPs. Not their whole sales process. Not every department. One high friction, high stakes bottleneck: hundreds of hours spent per RFP, in an industry where speed and accuracy decide who eats.
Their agent pulls together NTT’s service data, client history, and market insight, and produces a full RFP draft in about 20 minutes. What used to take hundreds of hours now takes the length of a coffee break. But the humans didn’t step out of the process. They shape the final proposal, tailor it to the client, and decide what NTT can credibly promise. The result: more opportunities pursued, and a higher win rate.
Vivian Lee’s framing for how you actually build something like this is the one I keep coming back to:
“Essentially, you’re creating a digital employee from scratch. The most important step is defining the desired outcome of what the agent is supposed to deliver, what this digital employee is supposed to do.”
Treat it like a hire, not a bolt-on. You wouldn’t onboard a new employee without deciding what decisions are theirs to make and which ones still need you in the room. NTT built that in from day one: the cockpit analogy Lee and Mantia both used. A pilot doesn’t manually execute every task, and a pilot is never removed from the decisions that matter. Define, ahead of time, exactly where a human has to step in.
That’s the piece that was missing from their own boardroom back in Canada, well before that afternoon in Cupertino, and it’s still missing from most of the AI pilots I get shown today. Nobody had defined where problem one ended and problem two began. Everyone wanted to solve all nine problems on the table at once, because that’s what disruptive sounded like.
NTT solved one problem. Then, presumably, they’ll move to the second.
If you’re evaluating a candidate process this quarter, run it through three questions before you touch it. Is this actually the highest friction point, or just the easiest one to reach? Have you defined, in writing, the decisions a human still owns? And could you explain, out loud, to your board, why this process and not the other nine sitting on the list?
Lee closed the webinar with the line I want to leave you with:
“It’s very important that we feel like we are in control, that we understand what these tools do, that we are governing them rather than the other way around.”
Nine airline executives flew to Cupertino to explore what was next, and how to be more disruptive. The real problem wasn’t sitting at that table. It had been sitting in their own boardroom back in Canada the whole time, and a brand new EVP was the only one who identified it, out loud. Grow, grow, grow had built the arteries. Process, process, process had clogged them. Nobody in that room, myself included, had a name for what she was describing.
It has a name now. Whether that changes anything depends on what happens the next time nine people fly somewhere to talk about what’s next.
Picture that room. The same instinct kicks in fast, someone reaches for the fashionable word, whatever this year’s version of “digital transformation” turns out to be (agentic AI, deployed without anyone asking why). Half the table nods along.
Then someone new to the team asks the smaller question instead. Not what should we add. What’s already piled up in here that’s slowing us down, and are we willing to look at it before we touch anything else.
That’s the whole difference. Not more ideas. Not a longer list. One person in the room willing to ask what’s already there before reaching for what’s next.
What’s your organization’s problem one, the thing you’d actually fix before any AI tool comes near it? Hit reply and tell me. I read every answer.
The newsletter isn’t the conversation. The conversation is the conversation.
See you next Friday.
Best,
JT
Sources:
HBR Webinar Summary, How Agentic AI is Reshaping Competition, August 27, 2026. Presenters: Vivian S. Lee, MD (Harvard Business School / Harvard Medical School), Linda Mantia (board director, Alan, Liberty Mutual, Prove Identity, Saputo; former COO, Manulife). Moderator: Julie Devoll (Harvard Business Review). All quotes and the NTT Data case study are drawn from this summary.