It's been a while since I have been in the community. Hopefully everyone having a blast.. Got the opportunity to attend Milton Keynes Tech Week event hosted on Strategic AI about how the local city council are moving towards opportunities for innovation on it. Was able to share my card across while networking with the city council and edited my first reel on Instagram to celebrate the begining of this journey towards AI in the local tech community. https://www.instagram.com/reel/De6eG2TN3cM/?obrf=aGEzdWdrdG5jZzFp Profile: https://www.instagram.com/SSaratKS Love your thoughts on it and following.

I bought tools from two vendors that were both pitched as a one-time payment. They did not turn out the same. The first vendor (Artistly, CloneVoice, TalkingPhotos, VideoExpress) has been one payment so far. I bought the bigger package, and in my own use the usage has been unlimited and updates have been included. I'm still testing VideoExpress. The second vendor (ToonBee, ClipBee, InstaDoodle) went differently for me. I got a basic version. In my experience, selling what I made commercially needed a paid Pro upgrade. The credits were not unlimited, and some features were sold separately. Added up, it was not one payment. These are the questions I'd ask now before any lifetime deal. Is the Pro or commercial license a separate purchase? Are any features sold on their own? If credits come with it, are they capped? That list comes from only two vendors, so it is probably incomplete. How do you work out the real total before you pay? Has a lifetime deal ever cost you more than you expected?

Spending this week going deeper on the wildlife AI system instead of citing the topline number again. The part I keep coming back to is the confidence threshold: every image gets a classification and a score, and anything below the threshold gets routed to a person instead of logged as fact. Setting that threshold is the part nobody talks about. Get it wrong in one direction and you're back to reviewing almost everything by hand, which defeats the point of building the system at all. Nobody hands you a formula for the right number. It comes down to a judgment call: how much risk of a missed or wrong classification is acceptable before a human needs to look. I don't think there's a universally correct number here. What matters is picking one that's honest about how wrong the model still gets things. Curious how other people building anything with a human-review step think about where to set that line.

I went back through everything I wrote in September and tried to find the thread connecting it, because at the time each piece felt like its own separate rabbit hole. Turns out the thread was simple: check the claim before repeating it, then go find what's already running for real. Week one was about email drafting. I kept seeing the same pitch everywhere, that letting AI draft your replies is basically free productivity. So I went looking for the research behind it instead of the marketing version. One clinical documentation study found unreviewed AI drafts were wrong 42% of the time. That number changed how I build now. The rule I run on is plain: AI drafts, a human sends. No exceptions, no matter how good the model gets. Week two moved off claims and onto client work I can point to directly. Workflow automation projects have been clearing 8 to 20 hours of manual admin a week for the people running them, worth somewhere between £20,000 and £57,000 a year in recovered value, a range that keeps repeating across different engagements rather than one standout client. The number that surprised me most wasn't the money, it was how consistently that range showed up across businesses doing nothing alike. Separate from that, a wildlife conservation project cut manual camera-trap image review by more than 90%, putting over 1,200 specialist hours a year back into people who'd rather be doing conservation work than sorting photos. And on the outreach side, an AI business development pipeline produced 140+ personalised emails across two niches in under two weeks, the kind of volume that used to mean hiring someone just for that. Week three tried to put a shape around all of it, because "AI helps businesses" is too vague to check against anything. I landed on a five-stage maturity model: Stage 0, manual everything. Stage 1, AI bolted on with no integration. Stage 2, AI inside real workflows with a human still reviewing. Stage 3, agents that each own a recurring job with human approval at the decisions that matter. Stage 4, a self-managing operating layer. Most small businesses I talk to sit at Stage 0 or Stage 1, which tells you where the real starting line is before anyone promises Stage 4 results.

I've been seeing a term floating around AI circles this year: "loop engineering." First reaction — another buzzword. So I went and checked where it actually came from before repeating it. Turns out it's real. Coined by a well-known engineer earlier this year, building on ideas a couple of others were already circling. But here's what surprised me: the underlying idea isn't new at all. Researchers were studying "reason, then act, then check, then decide again" loops back in 2023, and the pattern traces back decades further, into old control-theory and feedback-loop thinking. So what's actually new? Not the technique. The name gave people a shared way to talk about something that used to get scattered across three separate conversations — prompt design, agent reliability, AI observability. Once there's one name for it, you notice it's really one discipline: how do you design the trigger that starts an agent, the planner that decides its next move, the verifier that checks the output, and — the part everyone skips — the stop-rule that tells it when to quit? That last one is the piece I've started paying real attention to. I found a case where a company burned through an entire year's AI-tooling budget in about four months, spread across thousands of engineers, before anyone capped individual spend. Nobody meant for that to happen. It's just what a loop does when it keeps running and no one designed an explicit stopping condition. Here's what I've taken from digging into this: when I build anything agent-like now, I ask the stop-rule question before I ask the reasoning question. Not "how smart can I make this," but "how does it know when to stop, and what happens if it's wrong for a while before anyone notices?" It's a boring question. It's also the one most people skip, and it's usually where the real cost — in money or in mistakes — hides. Still figuring out where the line is between a good-enough stop-rule and over-engineered guardrails. If you've built agent loops and have a rule of thumb for that, I'd like to hear it.