The year is 2024. A group of writers who thought they had a complete book proposal looks at the landscape of technology, curses, and starts outlining a chapter on AI. We all write about AI now, except for the bits where AI writes about us.
Progressive Delivery is “Delivering the right product to the right person, at the right time.”. It’s a set of practices, not technologies, and it relies on abundance, autonomy, automation, and alignment to increase delivery speed, accuracy, and joy. We have a whole website about it, and the website is a subset of the (hopefully forthcoming) book. But you can’t be trying to sell a book in this environment without addressing the GPT in the room.
I haven’t seen a technology take off this quickly since visual web browsing. One fall, you learn how to use the gopher web browser, and by spring, everything is Netscape and has animated gifs of construction icons.
That was the start of what old internet people called “The Eternal September“. The number of new users vastly outnumbered the previous community, and it changed the way everyone interacted.
We are in the September, or perhaps the October, of this round of Artificial Intelligence Use and Research. It’s new, and exciting, and accessible. My web browser is using Copilot, even though I didn’t ask it to, because everyone believes that what we are searching for is answers, not resources. It’s a reasonable thing to believe. I often do just want to know how many millimeters in 1 1/4 inches. It feels like everything has a chat interface, a data miner, and a black box algorithm, and that’s obviously how it works, right?
Maybe.
But Usenet existed before AOL. LLMs, OCR, image identifiers, predictive text, and chatbots all existed before OpenAI and ChatGPT. What’s new is not the concept, it’s the volume and resources, the abundance.
Anyone (of the computing class) can log in and ask Notion to come up with a meeting agenda, or ChatGPT to explain a technical concept in terms of baking pastries. Automattic wants to know if I want help with boilerplate on my website and Google Gemini is ready to help me search for things. And these tools are amazing at a lot of things we find hard or tedious. They absolutely save me time and energy, even for things as simple as correcting all the times I mis-spell “ducking”.
Ok, they can’t do math. They only know what we know. We can’t actually see how they work. But they are great at the things they can do.
For me, the most exciting thing LLMs can do for Progressive Delivery is tightening feedback loops. If getting feedback from our customers involves someone having to call and complain to tech support, or even post something angry on social media, that’s a pretty slow, lossy, inaccurate feedback system. We’re not only missing the positive signals, it takes forever for that information to make it to product. Then product makes a change, and the slow feedback system starts again. With the magic of AI, we can train a model to look for deviations in user behavior, match and bundle the customers, and be able to see that “almost no one in Minneapolis buys swimsuits in February”. Instead, maybe Austin is a better place for that. Or maybe Minnesotans do buy swimsuits in the winter, but only if they’re going on vacation, so we could be more specific about how we stock them. The shorter the feedback loop, the more accurately we can understand our users, the better we can be at delivering what they want and need.
I find it charming how we appear to be coalescing on 
as our visual shorthand for “AI assistance”. It does sometimes feel like sprinkling some magic on. And right now we’re at a place where it seems like there are only advocates and refuseniks, with nothing in between. But like almost every technology, as it matures and we get used to it, we’ll coalesce somewhere in the middle, where it’s a tool that we don’t think much about, that powers other parts of our life.