Darkshift

Manifesto

Trust is the bottleneck now

By Sandeep Panda and Syed Fazle Rahman

We’re Sandeep and Fazle. For over a decade we’ve built one thing: software for the people who build software. Hashnode grew into a blogging community used by millions of developers, backed by Salesforce Ventures, Sequoia India’s Surge, and Accel. Bug0 came next: AI agents that test web apps end to end, with Passmark, our regression testing framework, open source at its core.

Two companies taught us the same lesson from opposite ends. At Hashnode, the backlog grew faster than any team we could responsibly hire; good ideas aged out in the queue. At Bug0 we live inside the other half of the problem every day: test suites that pass while software breaks. Shipping was never the truly hard part. Believing what shipped is.

Then coding agents got good enough for real backlogs. So we wired them into a production line for ourselves: isolated sandboxes, each repo’s own checks, browser QA recorded to video, one pull request per repo, a human merge at the end. We pointed it at Hashnode’s backlog and Bug0’s, and it shipped. When Ramp and Uber published what their internal agent factories were doing, we recognized the machinery. Ours just wasn’t tied to one company.

A few beliefs run through everything we build here.

01 · Evidence beats assertion

An agent claiming a change works is an assertion. A recording of the change working is evidence. Every Darkshift pull request carries evidence, and the PR is the approval gate: nothing merges without a human.

Claim versus evidenceOn the left, a claim: a summary saying the change works, with nothing behind it. On the right, evidence: the diff, the checks that passed, and a recording of the change working in a real browser.CLAIM"IT WORKS"NOTHING BEHIND ITEVIDENCEchecksPASSqa.webmREADABLEBY A HUMAN
Fig. 1 — A claim asserts the work is done. Evidence shows it.

02 · Factories are defined by inspection

Manufacturing settled this a century ago: output you don’t inspect is defect inventory. Much of our industry is currently building very fast lines with no QA station. We started from the QA station.

The inspection stationUnits of work travel along the line through a QA station. Units that pass leave solid and go on to a pull request. A unit that fails is drawn hollow and is diverted rather than shipped.WORK IN PROGRESSQACHECKS + BROWSERSHIPPED AS A PRHELD BACK
Fig. 2 — Every unit passes an inspection station before it leaves the line.

03 · You should own your factory

It runs on your infrastructure, with your keys if you want them, and the engagement is designed to end: our forward-deployed engineers hand the line to your team. We sell the factory, not a dependency on us.

The factory inside your boundaryFor a forward-deployed engagement the factory is built inside the customer's own boundary, in their cloud or on their premises. The repos and the credentials stay on their side of the line.YOUR CLOUD OR ON PREMISESTHE FACTORYrepossecretsYOUR KEYS, YOUR SIDE OF THE LINEWE BUILD IT,YOU KEEP IT
Fig. 3 — The factory runs inside your boundary, on your infrastructure.

04 · Agents are stations, not employees

We don’t name them, don’t give them faces, and don’t pretend they’re colleagues. Claude Code, Codex, and Grok Build are interchangeable parts in a line whose quality comes from the stations around them.

Interchangeable coding agentsClaude Code, Codex and Grok Build each drop into the same socket on the line. The agent is chosen per task; the clone, checks, browser QA and shipping steps around it do not change.THE LINESOCKETCODEXGROK BUILDCLAUDE CODEPICKED PERTASK
Fig. 4 — The agent is a part you can swap. The line around it is the product.

What we won’t pretend: that any of this replaces engineers. It changes where their judgment goes. Specs in, review out. The teams that get the most from a factory are the ones with the strongest opinions about what to build next.

We run Darkshift on Darkshift, and on Hashnode and Bug0, every day. When you book a demo, the factory takes a task from your backlog and ships the pull request while you watch. That’s the pitch. We’d rather show you the line running than write another paragraph about it.

From the founders

Sandeep Panda

Co-founder, Bug0 · Building Darkshift

At Bug0, Sandeep architects Passmark, the open-source AI regression testing framework. He previously co-founded Hashnode and scaled it to millions of developers as its CTO. He wrote AngularJS: Novice to Ninja, co-wrote Jump Start HTML5, and thinks mostly about agentic testing and the infrastructure that makes AI-written code safe to ship.

X · LinkedIn

Syed Fazle Rahman

Co-founder, Bug0 · Building Darkshift

At Bug0, Fazle builds AI agents that test web apps end to end. He previously co-founded Hashnode and helped grow it into one of the web's largest developer communities. A front-end and UX engineer by training, he wrote Jump Start Bootstrap and Jump Start Foundation for SitePoint and has spent more than a decade building products for developers.

X · LinkedIn

Watch the line run on your repos

A demo is a working session: your repos, a real task from your backlog, a finished pull request. Book a demo, or read how the factory works.