Skip to main content

Ten years.Three eras.Same bar.

I have shipped production-grade software for ten years: first by hand, then with ChatGPT, now with agents that write most of the code. Each shift moved my work up the stack, from writing lines to designing systems. The bar for what reaches production never moved.

How I got here

  1. The start

    Neural nets first

    I have been passionate about technology since the first time I got my hands on a computer. That curiosity took me deep into machine learning and neural networks as an intern at JSM Solutions in Hyderabad, years before neural nets started writing code.

  2. 2016 - 2018

    Line by line

    Then I became a developer at Awign Enterprises, on the JavaScript stack, when code was still written mostly by hand. As its gig platform grew, I learned to build as if every external call will fail.

  3. 2018 - 2023

    Zero to one, at scale

    At the Office of the CAG (Indian Audit and Accounts Department, Government of India), I took several applications from zero to one, thinking about the product as much as the code. I built data dashboards and enterprise software at scale, and got hands-on deploying them to government cloud and on-prem servers.

  4. Late 2022

    The first shift

    I started using ChatGPT as soon as it was released, and it was the first shift in how I built apps. Instead of writing line by line, I generated snippets. When I hit a syntax error, I used the debugger to find the relevant area first, then worked through the error with ChatGPT. Finding the problem was still my job. ChatGPT handled the syntax.

  5. 2023 - Now

    Agents, skills, harnesses, loops

    I have worked as an independent consultant since 2023. In the last couple of years, the landscape has evolved much faster. An agent writes code. A skill teaches it a task. A harness sets what it can see and touch. A loop keeps it going until the checks pass. The engineering has moved out of the code and into the system around it, and so has my work.

  6. Now

    Around the agents

    The hard part is not the agents. It is the system they work inside: named roles such as orchestrator, implementer, reviewer and gatekeeper, and tiered autonomy. I extracted that layer into sisu, an open-source harness generator.

Up the stack

Working across many languages, I realized we are becoming language agnostic, and the job keeps moving up the stack.

  1. JavaScript

    The job title was the language.

  2. Frontend developer

    Then it was a layer: everything in the browser.

  3. Full-stack developer

    Then the whole app: UI, API and database.

  4. Full-stack plus cloud

    Then running it too: containers, servers, deployments.

  5. Systems thinking

    Every engineer is now expected to operate at staff level, and that takes systems thinking.

Each step made the language matter less and the system matter more, and the top step is where I work now.