The Indian Canvas #79: Who pays for AI to write into old hospital software?

Minicor sells to AI companies that cannot go live because their customers' software has no API. I asked whether that buyer exists in Indian hospitals. My view: the plain integration job in India is not empty and is partly served already. What I could not find is evidence for the narrower job, writing AI output into a closed desktop system its vendor will not open. That is a hypothesis, and this issue ends with a way to test it cheaply. I have not spoken to a hospital IT head or an AI vendor.

The company that sells to the blocked

Minicor announced a $3.8M seed round on October 14, led by Brad Gillespie at IA Ventures, with Y Combinator and Paul Graham taking part. Its pitch, in its own words, is that AI should be able to use legacy desktop software like an API. It says 70% of Fortune 500 companies still run on desktop apps with no API. That figure is the company's and I could not check it.

The buyer is the interesting part. In its YC launch post, Minicor says its customers are AI companies selling into healthcare, automotive, financial services and logistics. They sign a customer, then cannot go live because the customer's system of record, such as an EHR or a dealer system, is a Windows app with no API. Minicor says it charges only when data gets in, and that it has gone live with a customer three weeks after a proof of concept. Those are company claims too. The point for an Indian reader is the shape of the sale: the payer is the AI vendor who is stuck, not the hospital.

What India already has, and it is a lot

Indian hospitals are not short of people offering to connect their software. The national health authority's developer portal describes what a hospital, lab or pharmacy system must build to take part in ABDM: identify the patient, then make records discoverable and shareable. The same authority publishes an open-source ABDM wrapper to help integrators. Its partner page lists software vendors with integration completion dates, and the live ABDM dashboard showed 5,93,109 verified facilities on the Health Facility Registry when I read it on October 20. A facility being registered says nothing about what software it runs.

Private players are in too. One connector product, ABDM Connector, says it is trusted by 500+ providers, which is its own claim. On the AI side, Bengaluru's Superhealth launched an agentic hospital operating system, SuperOS, in its flagship hospital, as The Financial Express reported in March. The same piece names several other Indian efforts, among them Tulu Health and a St John's Research Institute project with H Company. B12 and Ospia also pitch hospital AI systems on their own sites. An open protocol for scribe-to-EMR exchange started on GitHub in January, though it is tiny so far. So "AI inside Indian hospitals" is crowded, and I would not call the market empty.

Where the gap might still be

Most of the above works with cooperation: a vendor that exposes an interface, or a hospital that buys a whole new system. The case Minicor targets is different. The hospital runs an older installed program, its vendor offers no way in, and an AI scribe or billing tool still has to put something back into it. A connector vendor's blog post says older Indian hospital systems often lack APIs and use proprietary databases. That is marketing from a seller of the fix, so I treat it as a lead, not evidence. I found no independent count of how many Indian hospitals are in that position, and no AI vendor saying it lost a deal over it.

There is a softer signal. A Healthcare IT News interview with Narayana Health's AI chief says deployments stall at workflow level, and a HIMSS report it cites says hospitals need more support on workflow integration. That fits the idea but does not prove it. Workflow integration covers many things besides legacy write-back.

For contrast, accounting is a place where I would not look for this gap. Tally publishes its own integration routes, and many tools already use them. The no-API problem is a property of a particular program, not of Indian software in general.

The idea, as a test

Payer: an Indian AI vendor with signed hospitals it cannot connect, for example a scribe, a coding or a billing-claims tool. Missing piece: a managed way to write their output into the hospital's existing desktop system, with a person checking a sample of what goes in. India adaptation: the work is mostly mapping many local software versions, so a regional vendor may know them better than a foreign platform. Pricing per successful write, as Minicor describes, would be a test assumption, not a number I can support.

The test is a week of calls. Ask ten Indian healthcare AI vendors how many of their signed or piloting hospitals are waiting on write-back, and which hospital systems those are. Ask ten hospital IT heads whether they would let a vendor install a client on a clinical machine. If most vendors say their blocker is something else, such as consent, training or price, drop the idea. If a few name the same two or three systems, a concierge version would be one person doing the entry for one hospital, before any software.

How it could fail

Hospital IT teams may refuse an extra program on machines that run patient care. A wrong write into a record is a patient-safety problem, not a typo, and someone must own that risk. Whether automating another vendor's screen is allowed by the hospital's software contract, and how patient data rules apply to a third party doing it, are questions for a lawyer, and I assert nothing on either. Hospital system vendors may also open their own interfaces, which would shrink the gap. And a foreign platform like Minicor may simply come to India itself.

Sources and photo credits

Minicor seed announcement, Minicor YC launch, IA Ventures, ABDM developer portal (hospital, lab and pharmacy systems), ABDM dashboard, NHA ABDM wrapper, ABDM partner page, ABDM Connector, Financial Express (hospital AI systems), B12, Ospia, MedScribeAlliance protocol, Healthcare IT News (Narayana Health), Medi4u blog, TallyHelp integrations. Cover photo: tangled electrical wires casting shadows on a textured wall by Maxim Nikiforov (@maxniki) on Unsplash, free to use under the Unsplash License. It gives no location, so it is illustrative only. This issue has no chart because I found no figures that I could chart honestly.