ReadSigil

The Signal · Issue 05 · India & Emerging Markets

Maharashtra and England are putting AI on the clerk's desk, with the officer still signing

Maharashtra launched MahaINDUS, built on Sarvam's Indus Cowork, for 2,500 officials. In England, Google DeepMind's PlanAssist cut planning officers' time on householder applications by 58% in pilots. Both keep a person on the decision, and both will be judged on use, not launch.

By Shashwat Kumar · · 4 min read · Analysis

Abstract generative illustration: stacks of document-like cards being sorted into a single lit tray
Original generative illustration. No company imagery.

Facts are attributed and linked to their sources. The analysis and opinions are the author's own.

What it is

On 28 October Chief Minister Devendra Fadnavis launched MahaINDUS at Sahyadri Guest House in Mumbai, mid-day reported. It is built on Sarvam's Indus Cowork, which Elets eGov reported is now live for 2,500 state officials in 22 languages, including Marathi. Its tools include GR Expert for government resolutions, an email summariser and an RFP bid evaluator. Data stays in India and is not used for training. Elets, citing Analytics India Magazine, put the two-year outlay at ₹11.26 crore. Fadnavis said Maharashtra is the first state to use sovereign AI this way.

On 31 October Google DeepMind reported results from PlanAssist, built with the UK government, Google Cloud and Faculty. In pilots with Barnet, Camden and Dorset councils, it cut officer time on householder planning assessments by 58%, and more than 70% of the reports it drafted were approved with no edits. Householder applications make up as much as 70% of planning officers' workload. A 12-month beta now adds 12 councils in three cohorts, for 15 in total, with plans to offer it to all councils in England in 2027. A qualified officer approves every assessment.

Indonesia is moving on the rules. Two presidential regulations, on a national AI roadmap and on AI ethics, are in the signing process. Deputy Minister Nezar Patria has warned that agentic AI in banking "could lead to fatal consequences if an error occurs".

Why it matters

Government back offices are where AI adoption can be measured: time per file, rework, appeals. PlanAssist publishes the most useful of these, the edit rate. It shows how often the officer disagreed with the draft. In my view a no-edit rate above 70% needs watching from both sides. It can mean the drafts are good, or that busy reviewers have stopped reading closely.

The best number in either rollout is the edit rate, because it shows how often the officer disagreed.

GR Expert addresses a problem every Indian institution knows: finding the current version of a rule among circulars and resolutions that amend each other. Banks have the same problem with their own circulars. For Maharashtra, the useful next step would be to publish usage and accuracy figures, as the English pilot has done.

What I'd watch

  • Whether Maharashtra publishes usage and accuracy figures for MahaINDUS.
  • PlanAssist beta results across the wider set of councils, including appeal rates.
  • Whether Indonesia's AI regulations are signed, and what they require of banks using agents.

Sources

  1. Maharashtra launches MahaINDUS AI platform, rolls out new services on MahaDBT and Aaple Sarkar portals mid-day, 28 Oct 2026.
  2. Sarvam signs enterprise agreement with Maharashtra; Indus Cowork goes live for 2,500 government officials Elets eGov, 1 Nov 2026.
  3. Speeding up planning decisions with AI Google DeepMind, 31 Oct 2026.
  4. Indonesia prepares AI roadmap and ethics regulations ASEAN Coverage, 30 Oct 2026.
  5. Indonesia urges strict AI risk mitigation in finance, healthcare Antara, 31 Oct 2026.