I graduated into software in the nineties and have worked through four confident answers to the question "how do we know if this is going well?" I didn't observe them from outside — I was certified in them, implemented them, and taught them.
Process maturity. I participated in CMM and CMMI appraisals in software factories in the early 2000s, as both assessor and as manager of projects under assessment. I know exactly what it feels like to be on both sides of that.
The project management orthodoxy. Earned value, PMO governance, the discipline of the plan. I've held a PMP since 2002 and led a chapter of the professional body.
Velocity and the agile metrics era. Story points escaping the team and becoming a management number, with everything that followed.
DevOps performance measures. Implementing delivery performance metrics based on the State of DevOps research in a large media technology organisation, and later in travel.
And now: has AI made us faster?
The argument:
Each of these waves was a real improvement on what came before. I'm not going to stand up and say they were all foolish — I chose each of them, and I'd choose most of them again in context.
But each followed the same arc. A genuine insight, expressed as a measure. The measure proves useful. The measure becomes the goal. The measure gets reported upward, away from the context that made it meaningful. And then it starts describing something other than the work. I'll show the mechanism concretely for each of the four, with what specifically broke.
Then the harder part: I'll argue the current wave is following the same arc faster than any of its predecessors, and I'll be specific about which early symptoms are already visible. And I'll make the case for what has actually survived all four waves intact — because something has, and it's the part nobody builds a certification around.
How this fits the theme: every one of these waves was, at bottom, an attempt to make judgement unnecessary. To find a number that would tell us what was true without anyone having to decide. None of them worked, and the reason they didn't is the most human thing in our discipline. That seems worth saying out loud in a year when we're being told judgement is the thing being automated.
Tone: reflective, not a rant. Each wave gets credit for what it got right before I say what went wrong. I'm implicated in all of them, and the talk says so.
A note on evidence: all four examples are from my own career, across multiple organisations and industries, most of them ten to twenty-five years ago. Nothing here depends on any employer's data. Where I reference published research — the State of DevOps work in particular — I'll credit it properly.
Interactivity: I'll ask the room at the start how many have been measured by each of the four, which usually produces a useful map of the audience. Two short "have you seen this?" checks during the talk. Fifteen minutes at the end for discussion, which I expect to be the best part — there will be people in that room who lived these waves differently than I did, and I want to hear it.
Key points — you will leave this session with…
The shared mechanism by which good measures become bad ones, described concretely across four decades of examples
A set of early warning signs that a measure has started describing something other than the work
What each of the four orthodoxies got genuinely right, and what's worth keeping from each
A specific, falsifiable prediction about where the current AI-productivity measurement wave is heading, that you can test in your own organisation
An argument for what has survived all four waves — and why it's the part that doesn't get automated