Manish Singhmanishsinghkumar.in
Case studies

Work

Systems where AI does recurring work and a human keeps the decisions. Each case shows what was built, how it runs, and the calls that kept it safe.

Live
AI-operated business · AI literacy and brain health

An AI-run course business

An AI agent operates a live course site in logged sprints while every business decision stays with me.

60sprints logged
31posts live
4paid courses
Read the case Visit unai-labs.com

Three calls that kept it safe

  1. The agent never touches money or logins. It will not enter credentials or place an order. Payment setup waits for me, even when that blocks revenue.
  2. The live site is the source of truth. Every sprint begins by comparing the real site with the log, so drift is caught instead of assumed away.
  3. Decisions are separated from tasks. Pricing, ISBN, naming and gateway activation sit in my backlog. The agent does the work; I keep the calls.
Live
Procurement · Self-publishing library

The Procurement Codex

A reference library that builds, checks and publishes itself, one source-to-pay step per episode.

15episodes
$0running cost
1publish gate
Read the case Open in the lab

Three calls that kept it safe

  1. Nothing ships without a self-check. Automation without a gate is just faster mistakes.
  2. Build the engine once. Shared components keep quality consistent across all fifteen episodes.
  3. Say what it is. Every page states it is generated and directional, and asks readers to verify against primary sources.
Demo in preparation
Consulting diagnostics · Private build

DISCERN

Turns stakeholder interviews into an engagement brain: problems, hypotheses, levers and value cases, each traceable to what people actually said.

4domain packs
4build sprints
0fees set by the model
Read the case

Three calls that kept it safe

  1. Fees are computed by rules, never by the model. Commercials must be reproducible. The model proposes scope; deterministic logic prices it.
  2. Every claim traces to a verbatim quote. Insights carry the exact interview words behind them.
  3. It is tested blind. A synthetic client with deliberately planted contradictions checks whether the pipeline catches what a partner would.
Live
Executive education · 10 interactive simulators

Simulators that argue with evidence

Ten browser simulators that let leaders pull the levers behind bullwhip, AI spend, agent governance and change saturation, each fact-checked against its sources before publishing.

10simulators
7simulators corrected
0data leaves the browser
Read the case Open the lab

Three calls that kept it credible

  1. Label the model honestly. Every app page states whether its figures are checked, single-source or produced by the model, and links the sources.
  2. Check the sources before publishing, not after. A single false statistic would discredit the whole series. One app still rests on a single source, and says so on screen.
  3. Keep data in the browser. Nothing a visitor enters is sent anywhere, so the tools are safe to use with real internal numbers.
Point of view
Workforce transformation · Framework

When should an AI-infused role cost more?

A way to decide which AI skills justify a premium on client work, and which are simply the new baseline.

4questions per skill
2kinds of AI skill: baseline and premium
Read the case Read the essay

Three principles

  1. Do not relabel everyone. Calling every role AI-infused inflates the rate card until clients stop believing it.
  2. Expect the premium to decay. Today's scarce AI skill is next year's baseline, so every premium needs a review date.
  3. Prove it before pricing it. Eligibility needs evidence of delivered work, not a certificate. Certifications are easy to collect and say little about what people do afterwards.