Skip to content

Blog

Notes from building AI that has to keep working.

Lacewing Technologies is an AI and software development company. Since 2019 we have built AI agents, automation and custom software for teams that then have to run them every day, long after launch. We write here about the decisions that determine whether those systems survive contact with real users: scoping, model choice, evaluation, handover — and the ones we got wrong.

Essays
4
Sections
3

Updated RSS

From the studio

Most AI projects fail because they were the wrong one

The expensive mistake is not picking the wrong model. It is automating a process that was never the bottleneck, and finding out six months later.

Strategy6 min read

In this piece

  • Failure is almost always a scoping failure, not a technical one. The system works; nobody needed it.
  • The most visible process is rarely the most costly one, and visibility is what usually decides the pilot.
  • Four tests predict success, and a process failing any one of them is a warning worth acting on.
  1. From the studio6 min

    AI voice agent vs call centre in India: the real cost

    Per-minute cost of an AI voice agent against a trained human agent, what each actually finishes, and the honest case for keeping people on some calls.

    AI Agents · Dipak Bhosale ·
  2. From the studio7 min

    What does a WhatsApp chatbot cost in India in 2026?

    Build costs, Meta's per-message rates after the July 2025 pricing switch, model usage, and the monthly running number almost no quote includes.

    AI Agents · Dipak Bhosale ·

Next step

Got a question these don't answer?

Ask it directly. We'd rather answer it on a call than have you guess from a blog post.