Sav Pushparajah's Linkedin Analytics

This fried chicken food truck found its starving crowd. Over 300 kids at a skate park, burning calories and working up an appetite. Parents like me, watching over them, willing to buy whatever it takes to keep the kids entertained—just so we can get some work done in peace. Well played, Crunchy Fried Chicken.

The great promise of AI was that it works 24/7 and never turns up intoxicated. Turns out, when AI does get drunk, it doesn’t sober up in a few hours. Rollbacks can take time. I had an agent running using GPT-4o that was analysing some important real estate documents — and it was giving me completely fictitious settlement dates. Thankfully, Claude had just come out of an ice bath and a meditation session, so when we switched over, the results were perfect. In my view, one of the key benefits of building on tools like Lindy and Gumloop is being able to switch between different models. Not only does it allow you to choose the right model for a specific task, but it also enables you to put the model on the bench if it turns up drunk and needs some time to sober up. Sh*t happens when things are moving so fast, what are you doing to de-risk?

How many domains do you own? The highest number I’ve heard is 35. I’ve got 6, but I’ve let go of a bunch over the years. Most business owners I know have at least 10 + An idea pops into their head and—bang—they buy the domain. But most of those domains just sit in their account. Nothing ever actually gets built. Until now. Until vibe coding. Your shower ideas could exist before you even leave the bathroom. It’s a wild time to be alive. The value of those with an engaged audience is about to skyrocket!

Your new users don’t have thumbs. A lot of businesses are finding that AI is actually increasing their total addressable market. Usually, that means automating business workflows they couldn’t before. But for some companies—like Linear— it’s not just more workflows. It’s more users. Their TAM used to be humans. Now it’s humans and agents. Wonder what the TAM slide in pitch decks is going to look like now 😉

Congrats to Leon Rebello! If you don't know him (or even if you do), this is worth a read 👇

Sometimes things don’t go your way. This morning, my daughter put on her soccer boots, ready for her game. The whistle blew and within seconds, one of the soles came off. She adjusted her stride so the flapping sole wouldn’t slow her down. I run to the car and get her sneakers, but the ref takes her out for illegal shoes. Then the rain came. It started bucketing down. She kept playing, essentially running in socks. But the team only had 8 players, and she didn’t want to let them down. I searched the car for masking tape—no luck. Another dad had electrical tape. Game on. As quick as an F1 tyre change, dads went super-dad, and taped the shoe back together. Good as new. Another retape pitstop at halftime, as the other shoe started coming apart. Sometimes, things get scrappy. But with grit and some masking tape you get back in the game and achieve some pretty great things.

Career reinvention. With the pace AI is evolving, most people I know are thinking about it. Unfortunately, it’s not as simple as shaving off your beard after eight years and calling it a fresh start. Although that’s probably one step to recapture your youthful curiosity. Reinvention often means going back to basics. Becoming an operator again. Getting your hands dirty. It’s f*king scary, especially if you’ve been working for a while, but that’s where growth happens. There’s no shortcut. For the last 12 months, I’ve been building AI agents, some for internal use, some for clients, some for home, but this is the insight that hooked me. In my previous roles, I built software that was deterministic. You knew the inputs, you could predict the outputs. BUT everyone had to follow the workflow and that sucks, because it means change or transformation. And change means convincing people to do things differently. Now, with AI, we can automate flexible, personalised workflows that fit how people already work.Less resistance. Less change management. Happier days. BUT - and this is important - because we’re using generative AI, we’re shifting from deterministic systems to probabilistic ones. That introduces new types of risk. I recently helped a client automate commission calculations from PDFs. No change to their system - AI extracted the data, whether handwritten or typed, and calculated commissions. But what if the AI misreads a figure? What if the model updates and the results shift - which actually happened in this case? That’s a missed or incorrect payment. In this new world, systems thinking is critical. You need to understand the big picture and how the interconnected parts influence one another. Yes, you can stack one LLM on top of another, but that increases test time, compute, and cost. Not necessarily a bad thing — but at what point does that cost no longer make commercial sense? I’ve reinvented myself many times over the last 20 years. But this time feels like the most significant — and, if you lean in, possibly the most rewarding. So build. Build for your organisation. Build for your partner. Build for your clients. Don’t wait for them to ask. Just build.