Skip to content

Named · The Helix Lab, our own work

Agents did the work. A person held the gate.

88,460 Australian searches a month mapped to Helix pages. Ten pull requests merged across five sites on 21 October 2026, each with a person's approval. Rankings, traffic and enquiries are still to be measured.

The story

How the work took shape.

Helix work · Studio · Explore / Execute

Stage reached: Ten pull requests merged; search results still to be measured.

Several websites, one founder, and a search market nobody had mapped. The work was too big for one person and too sensitive to hand over blind.

The situation
The Helix Lab's work lives across several sites: the firm site, Suhit Anantula's books and frameworks, Ethnobot, Agent Watch and the 9Q site. Buyers search for problems such as AI strategy consulting or why AI pilots fail. Almost no one searches for Helix framework names. Key firm pages were hard for search engines to read. Client proof is sensitive: some stories are named, some stay anonymous, and only some figures are approved for public use.
What we did
We set up an SEO desk agent, running on Grok Bot, to research demand with DataForSEO and Google Search Console, route each search to one best page, and write a page plan for every gap. A delivery agent turned those plans into tasks for Cursor cloud agents, which built each page as a draft pull request with tests. The desk agent then read every draft the way a search crawler does and checked it against the plan and the claim rules. Anything wrong went back as a numbered fix list. Suhit kept three calls for himself: which clients are named, which figures are used, and what stays anonymous. Crediting City of Mitcham by name was one of those calls. Nothing merged without his yes.
What changed
Search work moved from a backlog to a running loop: research, plan, build, check, fix, approve. The agents carried the volume. The judgement stayed with a person. Review caught problems before they shipped, such as planning notes left in page copy and a client name placed next to an anonymous story. Each went back, was fixed and was checked again.
The evidence and status
88,460 Australian searches a month mapped to Helix pages. Search data for research rounds two to five cost US$0.52. Ten pull requests merged across five sites on 21 October 2026, each with a person's approval. Rankings, traffic and enquiries are still to be measured.
The pattern
If you want AI to carry real work, design the gates before the agents. Let agents do the volume. Keep naming, proof and the final yes with a person.

Let’s find the starting point

Does this look like your organisation?

Tell us about the situation and what needs to move forward. We can work out a useful starting point together.

Start a conversation