Writing · 2026

A court is a hospital with no patient chart

If you want to understand why Indian courts are slow, don't start with judges. Start with a hospital.

Most health outcomes aren't decided in the operating theatre. They're decided earlier: in triage, in diagnosis, in whether anyone kept a record of what happened last time. A great surgeon can't save a patient who arrives at the wrong hospital, three weeks late, with no chart.

That's roughly what an Indian court is. A tertiary hospital with no triage, no patient chart and no lab integration, where everyone gets sent straight to the surgeon and then we blame the surgeon for the waiting room.

India has no shortage of brilliant legal surgeons. What it's missing is primary care.

Where the time actually goes

The standard story about judicial delay is about judges. India has about 21 judges per million people; the Law Commission recommended 50 in 1987. Fill the vacancies, the story goes, and the backlog goes away.

But in some states that have expanded judge strength, pendency barely moved. Some courts dispose of more cases than are filed and still don't get ahead.

When you sit in a registry, you see why. A lot of delay is clerical. A filing comes in with a defect: a page numbered wrong, a document missing, the wrong fee. It goes back to the advocate. That can take a fortnight. Cause lists get set in ways that don't suit the parties, so dates are missed. When a litigant changes lawyers, the case can effectively restart, because no one else knows its history.

None of that is a judge's time. All of it is a case's time.

Pendency measures the wrong thing

"Pendency," the number of cases waiting, is the statistic everyone quotes. It's easy to count and politically irresistible. It's also misleading.

Pendency tells you how much load a system is carrying. It doesn't tell you how well it's working. It's like judging a hospital by how many beds are full. Half of the cases pending in the lower courts are petty offences like traffic challans, sitting in the same queue as murder trials and constitutional challenges. A triage problem looks like a capacity problem when your only number is the queue length.

And the biggest single source of load is the government itself. By one estimate the State is a party to about 73% of matters before the Supreme Court, often appealing on points of law that are already settled.

If you only measure the queue, you'll only ever try to make the queue go faster.

Primary care for courts

What would primary care for a court look like?

Triage. Check filings for defects and jurisdiction before they consume a registrar's time or a hearing slot. Fix small defects on the spot. In the courts where jhana works, filing scrutiny that took between one and a half and five hours now takes about fifteen minutes, and catches more defects than manual review did.

A chart. Keep a running record of every case, every hearing, every filing, every lawyer and litigant, so nobody rebuilds the story from scratch. A docket is supposed to be the court's memory. Right now it's a filing cabinet that forgets.

Referrals. When a court passes an order, it often needs to reach a prison, a police station, a welfare department. Today the court's job ends at transmission, and things get lost in transit. That plumbing can be built.

None of this involves AI deciding anything. All of it frees a judge to spend more of the day on the part only a judge can do.

Who builds it

Here's a choice India gets to make.

One option is that each court builds everything itself. Courts are remarkably autonomous; each is effectively writing its own rules. But if every department builds its own tools in its own silo, you get fifty incompatible systems.

The other option is the one India already got right once, with UPI. The state set the rails: the standards, the data rules, the security and audit requirements. Private companies built on top, and competed. Neither a state monopoly nor a private one.

I think courts should own the grammar, and let the ecosystem write sentences. Court-owned standards and data; certified vendors building specialised tools on top; and some form of procurement guidance, so that a judge who is also deciding a complex tax appeal doesn't have to personally evaluate AI vendors on the side.

The part that excites me

Courts are slow to adopt technology. That's usually treated as a weakness. I've started to think it might be an advantage. A system that never fully digitised has less to unlearn. There's a kind of institutional plasticity that lets you redesign the process instead of automating a broken one.

We're early. Most courts still run on paper and goodwill. But for the first time, the primary care layer is buildable. And if you build it, you don't just get faster courts. You get a justice system that remembers.