Kanav Jain

What I believe

Core principles, and the work each one came out of.

I like building things. The distinction I keep returning to comes after something works: making it work better versus asking why it needs to work that way at all. A successful solution can make life easier while also making the arrangements around it harder to reconsider. I’m interested in the choices embedded in technical systems, the work people do to keep them functioning, and who gets to decide what they’re for. The most interesting thing we can build is the capacity to move beyond what already works.

The argument behind these positions is in What I Hold Invariant.

• I’d rather show a working version than win an argument for one

  A proposal asks a reviewer to imagine the risk. A running product shows what it is. I build the first version myself and put it in front of the people who have to approve and use it.

  • Doximity: I expected to lose the caller-ID argument on paper, so I proved it over a weekend and launched. Hospitals approved it after physicians were already using it.
  • NextConsensus: The first test showed no lift from the signal I started with, so the forecasts now rest on each decision’s own review record.

• A workaround is a signal. Making it permanent is the failure

  Improvising is how an early product learns. When a mature system works only because people keep making up for it, with nobody counting the effort, that is a design fault, not resilience.

  • Epic: It went live with every sign-off, and it worked only because the people around it improvised.
  • The Crumple Zone: My essays on who absorbs a system’s failures when its design lets the burden fall on them.

• I don’t blame people for what they had no power to change

  Whoever controls a system is the one to answer for it. Anyone expected to correct an automated decision needs the authority to change it.

  • Epic: After go-live, responsibility was split between Epic and the hospital, and problems fell through the gap.
  • Andwise: Automated analysis fed the decision; a named reviewer with the authority to change it signed off.

• I check whether a service leaves people more capable

  A service can meet a need and still leave the person less able to act without it. The question is whether the arrangement builds skill, options, and a fallback, or consumes them.

  • Transcarent: One shared system gave nurses and navigators the same facts, so the next case did not depend on one person’s memory.
  • The Crumple Zone: The essays ask when an arrangement develops the people inside it and when it only extracts their adaptability.

• I want expertise to inform a decision, not own it

  Specialist knowledge is necessary, and it is not a permanent title to decide for everyone else. The test is whether the people affected can question the judgment and change the terms.

  • Epic: Hospital staff knew the workflow; the escalation path joined their knowledge to a named owner instead of treating either as the final answer.
  • Andwise: Physician advisors shaped the contract review; the person signing the contract kept the decision.

• Success does not confer sovereignty

  Doing something well today does not earn an institution the right to decide what comes next. It can deliver real benefits and still close off the alternatives.

  • Epic: The software was installed and working; that did not settle who owned the next decision.

• Being heard is not the same as changing the outcome

  Consultation, transparency, and complaint channels can absorb criticism without transferring authority. The test is whether the challenge can reach the rule, the budget, or the person who can change it.

  • Fast Harm, Slow Repair: A correction that reaches the record but not the rule leaves the next case to repeat it.

• Being able to do something isn’t permission to do it

  What someone may do and what they can reach are different things, and I ask who gave them permission.

  • Ambit: Ambit lists which tools an agent can reach and which it has used correctly, without assuming it may use any of them.

• When cases don’t fit, I check the category first

  When real cases keep failing to fit a category, I look for the problem in the category first, and only then in the person.

  • Refract: When a source stops matching its record, Refract logs a dated event to examine instead of discarding it as noise.
  • NextConsensus: Each forecast freezes before anyone knows the outcome, so one that turns out wrong is itself a result, not something to explain away.

• I judge a system by what happens when it’s wrong

  People affected by a system need a way to point out errors and challenge decisions, and those challenges have to be able to change the rules. Not everyone has that power, and it’s expensive to use, so I also look at whether the system finds and fixes its own errors before anyone has to force it, and whether it keeps a record when they do.

  • Refract: Downstream systems receive what changed and when, so a correction reaches everything the error touched.
  • Fast Harm, Slow Repair: A protocol measuring how far a wrong output travels before the correction catches up with it.

• A choice you cannot afford to exercise exists only on paper

  A right means little if using it costs the person with the least power the most. I check whether people can really refuse, challenge, or leave, not only whether they’re allowed to.

  • Andwise: A physician can push back on a one-sided contract clause only if they know which clause it is. The Contract Analyzer flagged them before signing.
  • Doximity: Physicians could technically choose any caller ID; the protection came from the account-level checks, not from the choice itself.

• Adaptation should widen what is possible, not excuse what stays the same

  People learn to live with arrangements that shouldn’t be this hard, and others often read that skill as proof it’s working. Adapting should leave people with more choice and more power, not become the reason nothing has to change.

  • Transcarent: Routing delays stopped being the bottleneck only after every case had a named owner; the workaround had been carrying the load.

I’d especially like to hear from people exploring these questions through research, engineering, political philosophy, or institution-building — including those approaching them from different starting points.

Woodturning and pottery

Outside software, I work at the lathe and the potter’s wheel, where wood and wet clay show a mistake right away. I built three interactive 3D labs for turners and documented pieces made in Chicago at Woodturning and pottery.