Harshal Gajjar

Harshal Gajjar is an AI Forward-Deployed Engineer at C3 AI, based in the San Francisco Bay Area. Harshal leads Agentic AI harness development for the Forward-Deployed Engineering organisation at C3 AI, and since January 2026 has been building a stealth-mode startup in the Agentic AI space. Harshal cofounded Shram.io in 2024, where he led the pivot from a Jira-competitor product to an AI assistant that reached #2 Product of the Day on Product Hunt.

Harshal holds an M.S. in Computer Science (Machine Learning specialisation) from Georgia Tech and a B.Tech in Computer Science from IIT Dharwad, where he was part of the institute's foundational class. He spent three summers at Wolfram Research in Boston — first as a summer researcher in 2018, then as an instructor for high-school students in 2019 and 2020 — and was a Wolfram Student Ambassador throughout his undergrad.

Outside of work, Harshal is a long-distance cyclist and a vertical and horizontal caver, active with the San Francisco Bay Chapter (SFBC) grotto. In 2019 he was part of the Hubballi Bicycle Club Guinness World Record for the longest single line of bicycles.

Contact Harshal at mail@harshalgajjar.com.

←Harshal Gajjar/blog
← map

blog · notes & longer pieces

a quiet feed — things i've been thinking about, mid-flight.

  1. piece·
    #

    Harnesses, not agents

    What "agentic AI" actually ships in the wild.

    Most of the production agent work I do isn't really about the agent. It's about the harness — the boring stuff around the model that makes a thousand customer-shaped failures recoverable, observable, and cheap.

    A model that's 95% correct in a notebook becomes 0% deployable without a harness that can: time out a runaway tool call, retry an idempotent step, fan out work, surface intermediate state to a human, and roll back side effects when the plan changes mid-flight.

    The interesting research question of 2026 isn't "can the model do it." It's "what's the smallest, most legible scaffolding under which the model can do it the same way twice."

    #agents#fde#c3
  2. note·
    #

    back from a TAG weekend. neversink in the rain — 162 ft of free hang, water sheeting down the cliff face the whole descent. nothing else in the week feels heavy after that.

    #caving
  3. piece·
    #

    Reading my own notes from Vipassana

    I sat a 10-day silent retreat in 2025. I came home and wrote nothing for a week, and then I wrote too much, and then I forgot most of it.

    The line I keep coming back to a year later: most of what I call "thinking" is reaction with a longer time-constant. Watching the reaction without grabbing it doesn't make it go away. It just stops compounding.

    The work hasn't stuck. The pointer has.

    #meditation#2025
  4. note·
    #

    8+ years into Mathematica and I still find a built-in I didn't know about every week. the language is fractal — there's always one more layer of "wait, that's a single function?"

    #mathematica
  5. piece·
    #

    The pivot is the product decision

    Notes from leading the Shram pivot, written down so I don't forget.

    At Shram in 2024 we were the 800th Jira competitor. The pivot — into AI assistants when the space was very new — was the entire company.

    Three things I'd tell past-me:

    First, the pivot isn't a feature change. It's a customer change. If your old user can still use the new product, you didn't pivot, you added a tab.

    Second, the speed of the pivot matters more than its accuracy. We had two weeks of runway, not two months, and the pressure forced clarity we'd been avoiding for half a year.

    Third, the team you pivot with is the team that decides whether the pivot survives the next pivot. Worth more than the idea itself.

    #shram#founder
  6. note·
    #

    old la honda → west alpine → pescadero loop. 78 mi, 6500 ft. the descent off haskins is the closest a road gets to feeling like a pipe.

    #cycling#bay-area
  7. piece·
    #

    What "forward-deployed" actually means

    A forward-deployed engineer is not a consultant. The job isn't to draw an architecture diagram and leave. The job is to embed in a customer's codebase, ship the thing that has to ship this quarter, and walk out leaving behind code their engineers will not delete the day you leave.

    The hardest part is restraint. You can see the ten things you'd refactor. You ship two of them and write a doc about the other eight. The two you ship are the ones that have to be there for the next FDE to land softly.

    #fde#c3
  8. note·
    #

    reading: "The Beginning of Infinity" (deutsch, again), "Designing Data-Intensive Applications" (kleppmann, for the third time), and the SWE-bench papers. one book per attention budget.

    #reading
  9. piece·
    #

    Sixteen years of writing code

    Today I'm sixteen years into writing code. The first program ran in Visual Studio in 2010. I was thirteen. It printed a number and felt like a magic trick.

    The trick has gotten more elaborate. The feeling hasn't changed.

    #origin#2010
  10. note·
    #

    first trip into Guardian with the SFBC grotto. tight breakdown, then it opens into a room you don't expect from the entrance. cave geometry is the best argument against intuition i've ever found.

    #caving#sfbc