Okay. Let me just say this properly.
So the newsletter crossed 1,000 subscribers this week.
1,000 is nothing in newsletter world. People do that in a weekend with one lucky post. I know that. But those aren’t the thousand I got.
The thousand I got are people building AI inside banks, insurers, healthcare companies, startups, consultancies. People who reply to my emails saying “we hit exactly this last quarter.” That’s a different thousand. I’d take it over 50,000 randoms any day.
You give me space in your busy inbox - that matters a lot to me.
Thank you for that.
Why I started the newsletter?
Everyone’s shouting about AI. Demos everywhere, announcements everywhere, everyone’s an expert on LinkedIn by Tuesday. But almost nobody talks about the boring part - what happens six months after the demo, when the thing’s in production, the data’s messy, the regulator’s asking questions, and someone senior wants to know why the numbers moved.
That part. The unglamorous part. That’s the whole newsletter. Because that’s where I’ve spent my career, and honestly, that’s where yours gets made too.
So nothing changes at 1,000. I’ll keep writing about the stuff that actually breaks - evaluation, data debt, decisions nobody wrote down. I’ll keep sharing the failures, mine included, because that’s where the lessons are. And I’ll keep reading every single reply, because half my best issues started as one of you saying “we’re seeing something weird and nobody’s writing about it.”
Keep sending me those. Seriously.
One more thing. I’m not going to take up space in your inbox just to celebrate - that’s not what you signed up for.
So here’s something useful. I just published a video on why most AI projects fail. It’s everything I’ve learnt working with enterprises over the past few years - the patterns I keep seeing, and what I think actually makes these projects succeed.
Watch it here.
Talk soon,
Sandi
P.S. If you’re new here - welcome 🎉. AgentBuild is a community of practitioners working through the real challenges of getting AI into production inside large organisations. Every week I share practical, grounded thinking from the people doing this work at the sharp end. The goal is never theory - it’s always: what can you use Monday morning.
Ask your friends to join.
More valuable content coming your way.
Congrats on the first thousand, Sandipan - well earned. “Beyond the hype” is the whole game. The way I keep seeing it: most AI that demos well is a read system, where being wrong is cheap and you just rerun it. The systems that truly work are the ones that have to act - and the moment an agent acts in the real world, a wrong move costs something you cannot rerun. That gap is where most pilots quietly die. Here’s to the next thousand.
Congrats on 1,000 - and the thousand you describe sounds like the right one to have. The line that landed for me is the boring part after the demo, because that is exactly where we spend most of our time too, and almost nobody writes about it honestly.
One thing I would add from the calling side of that same problem: the boring part is not just evaluation and data debt, it is also that a lot of production AI now takes actions in the physical world - a call placed, a message sent - and that changes what testing even means. You cannot diff two live phone calls the way you diff two API responses. Curious whether you are seeing that shift show up in the enterprise agents you write about, or if it is still mostly a text-and-retrieval problem for most of your readers.