AI-augmented product practice
Strategy, research, design, and code no longer have to wait their turn. AI gets a team from a premise to a working artifact faster. Leadership sets the outcome, reads the evidence, and decides what the next pass needs to prove.
Live agents watch a few things I care about. They speak when a connection becomes timely, useful, or strange — and go quiet when it does not.
AI timeline
Short dispatches on where the technology is heading and what it changes for building products. One card at a time — arrows, click the center or right card to move forward, click left to go back, or scroll sideways.
June 2026
The loop is starting to close on itself: build, test, inspect, repair, with less of a person in the middle each month. What I’m watching is where it should stop — which calls stay human when the machine can technically make the pass, and how you keep any of it pointed at real customers instead of its own scorecard.
April 2026
Attention moved off the model and onto everything around it — what context it’s handed, what tools it can reach, where a person still signs off. There’s a name for it now, context engineering, and it marks the moment the model became a component instead of the product.
February 2026
Evaluation became what everyone talks about. Teams started writing success down as checks they can actually run, rather than settling it by opinion in a review. It’s the first real quality bar I’ve seen for systems that don’t give the same answer twice.
November 2025
Agents left the chat window. The model stopped waiting for the next instruction and started running its own short loops — try, check, adjust. The question changed from what a model can say to how much of a task it can carry before it needs a person again.
July 2025
The work shifted from describing things to standing them up. Instead of a deck about an idea, you could put a working version in front of people and let them push on it. A room arguing with something real, instead of imagining it, was the tell that the ground had moved.
March 2025
Everyone learned to prompt this spring. The field filled with technique: personas, phrasing, step-by-step reasoning, all of it treating the model as the whole machine. I kept thinking the real leverage would end up somewhere the wording of a request couldn’t reach.
November 2024
Generation stopped being the hard part. A credible first draft of almost anything is seconds away now, and the models crossed from novelty to dependable inside a single year. The bottleneck moved — and most teams were still optimizing the part that had just gotten cheap.
Where AI helps, and what people still own
AI opens more directions and makes working artifacts cheaper to produce. It does not define the problem, understand the promise the brand has made, or decide which answer deserves investment.
Working with AI means resisting the false relief of the first good answer. A plausible answer can arrive quickly now, which makes it easier to confuse momentum with quality.
The discipline is knowing when to stay with the problem a little longer: test another direction, sharpen the criteria, combine what almost worked, and do not accept output simply because it looks polished. The opportunity is not to generate endlessly; it is to keep the work open long enough for a better answer to emerge.
In practice
At Arity, I used this workflow to make new-market opportunities tangible. The desktop prototypes gave product, engineering, analytics, sales, and prospective partners the same artifact to question together.
The payoff is not an AI demonstration. It is a team that can see the same product direction, challenge it with evidence, and decide what to do next.
Back to Leadership