We hallucinate too
We worry about whether AI makes things up. We rarely ask whether we do — rewording, regenerating, or filling the gap until it tells us what we wanted.
Engineering · Travel · Photography
I'm Manohar. At an energy-tech startup Racanaa in Bengaluru I look after engineering across the board — the full software lifecycle, data, QA, security, DevOps, product support — and, it being a startup, the IT and infrastructure too: networks, firewalls, keeping the whole thing standing. 'VP, Data & Engineering' is the title on paper; in practice it's all of it. These days I'm playing with AI almost every day to build and test new products. I've been making photographs since 1996 and I'll take any excuse for a road trip. On the side, my wife Roma and I are reviving our ancestral home in the Almora hills as Kunja Village Homestay. My travel and photo journal lives separately at mnegi.com.
We worry about whether AI makes things up. We rarely ask whether we do — rewording, regenerating, or filling the gap until it tells us what we wanted.
We've been trying to actually run an AI-driven development lifecycle — a real process the AI follows, not prompt-and-hope. Here's the honest state: what we've learned, what we're really after, the gap we keep hitting, and the options we're weighing.
AI made writing code nearly free. The value didn't vanish — it moved to what was always the real job: judgment, architecture, and holding the bar in review.
Not toy prototypes — shipping. The working loop, habits and guardrails I use to build with AI every day, and where my own judgement still has to take over.
A clean, content-first Next.js engine: MDX files, a tiny taxonomy, static search, GitHub-backed comments, and likes that survive a redeploy. A tour of the moving parts.
No schedule, no spam — just new essays when they’re ready. The slow web, delivered slowly.