AI and the Blurred Lines of Product and Engineering

Let's cut to the chase: the line between product and engineering is getting fuzzier than a cheap sweater. We're not talking about a gentle merge, but a full-on collision, thanks to our new digital overlord: AI.

Hit the rewind button and cruise through the annals of history with me. Picture the 1980s: software development was drenched by the Waterfall model, a bygone era where product development was a straight shot from A to Z. Back then, the rules were simple and the boundaries were ironclad. Engineers and product managers operated in their own silos, passing deliverables like batons in a relay race. It was orderly, it was predictable, and it was safe.

It was also a bit dull. The Waterfall model was the trusty old typewriter in a world that was about to be rocked by the personal computer revolution.

As we moved into the 2000s, the Agile Manifesto flipped the script. Out with the old, in with the rapid-fire evolution of products, sprint by sprint. Enter the World Wide Web, the ultimate disruptor, forging unprecedented connections between users and devs.

As the clock struck 2010, the DevOps rebellion kicked in, smashing the barriers between development and operations. The rise of cloud computing and relentless CI/CD practices sparked a revolution, enabling an anarchic rhythm of updates. Products evolved at a breakneck pace, with feedback loops so tight they'd make your head spin.

The 2010s paraded in with a Big Data and AI fanfare, pretending to be the saviors of product development with their predictive analytics and machine learning gimmicks. But let's be real, the so-called 'data-led' revolution often feels like a wild goose chase in a hall of mirrors.

And now, here we are, teetering on the edge of a brave, chaotic future where AI is breaking the wheel, and it's not stopping there. It's slashing the tires, hotwiring the engine, and taking us on a joyride through uncharted digital territories. Buckle up, buttercup – it's about to get wild.

The AI-Infused Product Lifecycle

Where are we now? In my current role, we're busy experimenting with GenAI in product specification documents. We've been pretty successful, with relatively little effort. Download a zoom conversation, pop it in your folder, whip up a few prompts about what you're looking for, and lean back while GPT makes you question all the effort you've put into these documents in the past.

Elsewhere, our engineers were already fed up with our specifications, and they're using AI to take a task description and whip up a nice, medium-rare implementation plan, so they never have to read our brilliantly written specifications again. Soon, they'll be deciding how to separate the tasks as well (oh, the horror).

Where does that leave my precious product backlog?

Gone are the days when product teams would conjure up grand visions and toss them over the wall to engineering, covering their ears to hide the explosions happening on the other side. Now, we're seeing AI bulldoze its way into our development cycle, and it's changing the rules. We're not writing the specifications for humans anymore, we're writing them for the AI. Do we even expect our engineers to read the specs anymore?

(did we ever?)

In the end, it's just another revolution in the product lifecycle. Adapt or die, as the kids say.

Dreaming Up the Future

Personally, I can't help but dream whenever AI is involved. The possibilities are supposed to be limitless, so let's at least aim for the stars.

Currently, we're here:

Frankly, that's an obscene amount of effort.

I've tried pitching the idea of managing a video call based around AI. We've put a lot of effort in our team into figuring out the best file formats and styles for giving an LLM text data. What do I want next? I want a video calling service which manages your meeting around the output you want to give the AI.

It should allow you to structure your meeting with controls in the UI. If someone wants to bring up a new feature idea, there should be a control that says 'New Feature'. Now, while we are talking there should be a button that says 'Use Case' and 'Decision to Document'. When the transcript is generated, the extra context should help the AI structure the entire conversation into a pre-defined format, and kick off automations which add this documentation into your repository, with a PR than can be reviewed manually by the product team.

After that gets merged, the product specification should be converted into an implementation plan, and then a version 0 of the acutal implementation in a PR for the product repository. An engineer then fixes the implementation, and we are ready for go live. What could go wrong?

The actual flow should be this:

• product meeting
• black magic
• quality control
• ???

I think you know the rest.

Anyway, let me get back to work. These product specifications aren't going to write themselves.

Yet.