Every few months, I used this space to talk a little bit about my day job. It seems like that post always includes a link to Benedict Evans's latest essay: in July I teased the current state of the AI discourse and noted his takedown of OpenAI's business model. It's been two months, and the AI discourse is in about the same state, with the executives responsible for the latest and greatest AI advancements crowing for the sort of regulation that prevents them from doing bad things by ensuring they maintain their leadership in the space. (Beware the industry leaders begging for governmental control—their goals aren't altruistic and whatever is done will certainly preserve their marketshare and likely result in the ongoing subsidization of their businesses.) As a software company employee, these sorts of considerations seem to me disconnected from what really matters. What's actually changing people's work with software is, thus far, more marginal. Most people aren't becoming software developers nor pushing fancy AI models to their outer limits nor even getting really good results from whatever workplace computer programs are latest on the scene. Most people use little bits of software here and there to support their actual work or to log their actual work. They might be after slightly easier or slightly less frustrating interactions with those software, but most people aren't excited about big changes nor do they want to take on seemingly big risks for potential returns. In the words of Geoffrey Moore, most people aren't early adopters of new technology.
I work at a tech company. We're the epitome of early adopters. We get so excited about a new way of doing work or a new tool or an adjustment of an existing tool, that we try the new way before stopping to assess whether the old needs to be improved. As you'd expect, I've had easily six different "daily driver" AI tools since 2023 and am about to make another big switch. While in sum all of these changes has been moving towards technically better tooling and the work I deliver has gotten better, the path hasn't been linear and the "but make it actually work" period for each tool has resulted in, temporarily, a step back in my ability to do good work. That's what the early adopter likes: spending as much time on getting the new setup in place as doing the actual job. Even in software-heavy jobs, early adopters are maybe twenty percent of the people. Most people hate the thrash of constant software upgrades and most people are annoyed, not excited, by having to learn a new interface or process in the slim hopes of slight improvements. Most people want someone else to go first.
Since October of last year, I've worked as in a "forward-deployed" role running what we've called "AI transformation" projects with our customers. The "forward" part of this work has meant identifying what business processes or outcomes need to be improved and then hunting for ways that ours or other vendor's AI tooling could deliver the needed performance or improvement. The "deployed" part of the term has meant that instead of just making recommendations, I've done some of actual work—building, testing, and launching of the new tools or process. All kinds of lessons have emerged from the work. Most are based on the premise that while people might want lower effort or better results from their tools, the challenges to arriving at those destinations are rarely about the tooling itself and far more often about the people, meaning their approach to work, their risk tolerance, and their openness to change. As I've learned and re-learned this lesson, my discovery work has focused on far fewer areas, I've become more and more attuned to who is making decisions and how change might happen (or not), and become more likely to have people to do the work with me.
At a higher level, the work of AI transformation, while it might seem different from what's happened before because of the technology at play, isn't really any different from any other kind of technology adoption work. Some people might get excited by meeting summaries, automatic slide deck creators, or chat being new universal user interface. These are the types of early adopters who believed that email and personal computers and ubiquitous smartphones were going to really change things. No one would say that things weren't changed: in the last forty years, email, computers, and smartphones became part of how almost everyone does work. But simply using technology doesn't improve processes or result in good outcomes. We're all familiar with meetings that could've been an email and with endless email threads that should've been a conversation. Positive transformation takes far more focus on what matters than on what nifty new tools can do. And, for me, that's where the work gets exciting.
Reading
AI, tools and transformation
It's very tempting to imagine that AI turns everyone into a tool-builder - now everyone can just ask the model to make the software they need, and apps as we know them are dead. I think that misunderstands how most people think and where software actually comes from, and more importantly, it isn't a path to change how companies actually work.