Last week, I ran the first Product Talk meetup, with the banner of “From Prompt to Product”, here in Melbourne. After spending all day wondering if anyone would turn up, if the drinks would be cold and how I could keep the Pizza hot, people did arrive, and I tried pretending to be cool about it all and that I hadn’t been panicking all day……
We had two speakers from two very different worlds, but both with their thoughts on AI and how they’d been using it.
Joel Kenyon, Former Head of Product, Amused Group
Joel has spent over a decade in product, most recently leading the design and delivery of a wagering platform across 10 brands. Not a startup playing in a sandpit. This role was about real compliance constraints, real customers and real consequences. When Joel talks about embedding AI into a product organisation, he means a fully functioning, established product organisation and all the challenges that come with it.
- His first big takeaway was a taxonomy of GenAI behaviour that was genuinely funny and also a bit too accurate. He called them the GenAI delusion archetypes. There’s the kool-aid drinker, who spends five hours prompting a monthly task that takes 15 minutes and then writes a LinkedIn post about how they changed the world. There’s the sceptic, who spends five minutes half-heartedly prompting the most complex thing they do, concludes it doesn’t work, and tells everyone at work that GenAI is a fad. And there’s the bandwagoner, who tells everyone “Something Big is Happening” but cannot give you a single real example. If you didn’t see yourself somewhere in that list, you probably weren’t paying attention.
- His second takeaway was about what AI actually did well in a real product workflow. He started using it to review functional requirement documents, training it on a mix of early drafts and polished final work, covering different types of artefacts, and narrowing its focus rather than letting it roam. The results were real: turnaround time dropped, quality improved, and review conversations shifted to substance rather than formatting. The process then scaled to other artefact types. It wasn’t glamorous, but it worked, and that’s the point.
- His third takeaway was about honesty. Joel gave a clear list of what hasn’t worked, and this was arguably more useful than anything else he shared. Strategy. Writing documents from scratch. Anything where the required knowledge isn’t readily available online. Complex research with real consequences. Shippable code. The point wasn’t that AI is useless; it was that people waste a lot of time expecting it to do things it isn’t suited for.
Starting with the right use case matters more than just starting.
Jared Korinko, Product Manager, Google Photos
In his normal job, Jared is operating at a scale most of us will never deal with. Over a billion users. He also came in with what felt like genuine intellectual restlessness about what AI is doing to the whole game of building software products.
In his spare time, he’s built a fully operational business with real customers and real revenue. You can check it out here – https://www.withsammy.ai/ He opened by saying he was scared, and then immediately said he was also having more fun in tech than he’d ever had. That tension turned out to be the thread running through his whole talk.
- His first big takeaway was a side-by-side comparison of what it used to take to build a SaaS product versus what it takes now. Before 2023, the process involved 12-odd painful steps, a developer mate who was always too busy, a dodgy offshore development company with questionable invoices, nine months of waiting, and ultimately wondering why you didn’t just buy Bitcoin. In 2026, you have an idea, ask an AI tool to build a prototype, wait seven minutes, demo it to real people the next day, and start selling. The compression of that process is not a small thing. It changes what you can build, how fast they can learn, and what they’re willing to try.
- His second takeaway was a line worth thinking about. Jared said that building an AI product means you’re building a “human being simulator”, and if you’re not an expert in the domain you’re simulating, you’re going to come unstuck. It sounds obvious when you say it out loud, but to me, it cuts through a lot of the noise around AI replacing expertise. The system prompt, the framing, the constraints you set, all of it depends on someone who actually knows how that domain works. The model isn’t going to figure that out for you.
- His third takeaway was about measurement. The model is the product, and you cannot improve what you do not measure. You need to be logging and reviewing traces daily (reading through a sample of those interactions to see how the model is behaving in the real world. Is it answering the way you intended? Is it going off script? Is it making mistakes in a particular type of situation?), grading and classifying them, and building evaluations over time.
If you’re not doing this, you are, in his words, building blind.
- This feels like the kind of operational discipline that separates people who are actually building from people who are just prompting.
On predictions
Jared wrapped up with four predictions for 2030 that are worth sharing.
- Unicorn companies with fewer than five employees will become a real thing, not a curiosity – Historically, this is impossible. You needed an army of sales reps, a legion of engineers, and a floor of support staff.
- SaaS pricing as we currently know it is probably dead – Usage-related billing (pay per credit) is quickly becoming the norm.
- Most internet traffic will be generated by AI agents rather than people – Agents will become the end users of the internet, using it to transact, communicate, and ask humans for help where physical bodies are needed.
- Proof of humanity verification will become commonplace as a result – It’s only natural we’ll want spaces on the internet for just humans, so proof of humanity will become commonplace to participate in these spaces.
Whether all four land is anyone’s guess, but I dont think they’re that wild 🤪 Maybe they’re just logical extensions of where things are heading.
A final note
The event ran at the office of our good friends at Ippon Australia and it was a brilliant space for it. Good room, easy to find, accessible, and the team were generous in making it available. I genuinely appreciate the support, and we’ll be back in the future.
Obviously, a huge thank you to Joel and Jared for giving up their time and to everyone who joined us on the night.
I’ll start planning the next one soon. So if you have something you think is worth sharing, or know someone who does, drop me a message.
Before you go, can you help me out?
Now over 1300 people have subscribed to this newsletter, which makes me feel that it’s resonating. So thank you to all of you 🙏🏻
If you found this issue useful, here are three ways you could support me.
- Share it – Forward this to a founder, leader. hiring manager, or job seeker who you think might find it useful and/or interesting.
- Follow and/or connect with me on LinkedIn if you’re not already – https://www.linkedin.com/in/simonfromcrew/
- Send feedback – What topics would help you most? What’s working? What isn’t? Just hit reply.
These small actions make a massive difference in helping me reach more people who need this content.
Thanks for being here. Have a brilliant week ahead.
Weekly newsletter drop: Every other Wednesday, 1:30 pm
Need hiring help?
If you’re struggling with a specific hiring challenge, I still offer 1:1 strategy calls through my consultancy, Crew. Book a 15-minute call by emailing me at simon@wearecrew.io
