Skip to main content

Lab Notebook · Chat Extension Line

Five versions in a month — the Chat Extension line and the habit it started

Across July 2023 we shipped five iterations of a browser-based LLM chat tool in one month — three Chrome/Edge extensions and two standalone apps, moving from vanilla JS to a Node.js backend with local storage and an API connector. None of them shipped as a product, but the pace itself — five versions in thirty days — is the direct ancestor of the far faster iteration habits that came later.

JR
Jon RossFounder, vocabotics — 15 years building safety-critical systemsLab report · dated 20 July 2023
Verified by a human. Drafted with AI, verified by a human. Jon Ross, 20 Jul 2023
Living document. Reviewed 20 Jul 2023
Entry date
20 July 2023
Category
Tooling
Lead over the world
the habit, not the lead
Access
🔓 Public
ResearchDisproven

Five versions of a browser chat tool in one month, in July 2023 — the direct ancestor of the '13 versions in 48 hours' pace AI tooling later made possible.

5
iterations shipped in one month (Jul 2023)
3 → 2
extension versions, then standalone versions
1
related novelty: BBC-GPT, a 1980s BBC Micro talking to GPT-3

Honest evaluation

Disproven

Judged as products the report calls this a null result in its own words; the rapid-iteration habit it seeded is the separate, surviving finding.

What would prove or disprove it further

What would prove or disprove it further: the report is honest that a different, secondary claim survives this verdict — the iteration pace itself, five working versions in thirty days, is credited as the direct ancestor of the much faster "thirteen versions in forty-eight hours" habit later. That secondary claim is falsifiable by tracing the actual project history of that later, faster era and checking whether the same throwaway-and-restart pattern genuinely shows up in it, rather than resting on it as a stated lineage.

The evidence — full reasoning behind the verdict

Verdict: disproven, on the terms the report itself sets. Judged as a product effort — did any of the five browser-chat iterations become something worth shipping? — the report calls this "a null result" in its own words, and most of the five were later archived outright.

The mechanism is simple: five rewrites in a month is a sign of restless iteration, not of a concept converging on product-market fit — each version fixed something the last one got wrong, but none of them found a reason to stop rewriting and ship. That's a real, self-measured miss against the implicit bar of "build a chat tool worth keeping."

Some lessons come from the code. This one comes from the calendar: five real iterations of the same idea, in one month, in July 2023.

What it was

A browser-based chat tool for talking to an LLM without leaving the tab you're in — built, rebuilt, and rebuilt again across July 2023. Three Chrome/Edge extension versions, then two standalone apps that dropped the extension dependency altogether.

What we built

Research

The extensions moved from a vanilla-JS front end to a small Node.js backend, picking up a FontAwesome/BeerCSS interface along the way. The standalone versions went further: v2 added local JSON storage and a proper API connector, so state persisted between sessions instead of living only in a tab.

In the same window, a related novelty: BBC-GPT, a Python proxy that let an actual 1980s BBC Microcomputer talk to GPT-3 through a BASIC front end. Different hardware, same instinct — put a modern model behind an unlikely interface and see what happens.

What we learned — including the honest negative

None of the five versions shipped as a product. They were prototypes, built to learn, and most were later swept into the May 2026 cull along with the rest of that era's debris. Judged as products, this is a null result.

Judged as a habit, it's a real one: five working iterations of the same idea in thirty days, each fixing something the last one got wrong, is a rapid-iteration discipline in miniature. That habit is the direct ancestor of the much faster "thirteen versions in forty-eight hours" pace the team reached once AI-assisted tooling caught up to the ambition. The tools got faster years later; the willingness to throw away a version and start the next one was already here.

Where it went / status

Archived. None of the individual tools survive as products, but the iteration habit they established runs, largely unbroken, all the way to the current operating model — just at a completely different clock speed.

What is still open — kept visible

The honest edges, next to the wins. This is what turns 🔬 into 🟢 — honestly.

  • None of the five iterations shipped as a product — they were all prototypes.
  • Most were later archived outright in the May 2026 cull.
  • The rapid-iteration pace was itself the finding here, not any one version's feature set.

Where this connects

Sources

  1. vocabotics project audit — Chat Extension v1-v3 and Chat Standalone v1-v2 (five iterations, one month), Jul 2023vocabotics internal project history · as of July 2023

    We use cookies.