The Discovery Journal: Unfolding Atlas

Atlas is not built behind closed doors.

The Discovery Journal documents the conversations, questions, breakthroughs, setbacks, and discoveries that shape Atlas as it unfolds.

Some entries are technical.
Others are philosophical.
Many begin as simple questions over a cup of coffee.

Together they form the living history of Atlas.

The Atlas Journey

Current Stage of Discovery

Atlas is being developed one conversation, one paper, and one discovery at a time.

Before writing code, we chose to build the foundations first. Over the past year we have developed the philosophy, governance framework, stewardship model, cognitive principles, and research that define what Atlas is and what it should become.

Alongside this work, we have completed multiple rounds of stress testing, explored human–AI symbiosis, developed the first Atlas decision algorithm, and begun defining the cognitive engineering and core architecture that will underpin future development.

Rather than rushing towards a prototype, we are taking the time to ensure the foundations are robust, transparent, and aligned with our core principles. We believe thoughtful engineering begins long before the first line of code is written.

The roadmap below provides a snapshot of where Atlas is today, what has already been achieved, what is currently being developed, and the journey still ahead. As Atlas evolves, this roadmap will continue to grow alongside the Discovery Journal, documenting not only what has been built, but what has been learned.

What have we actually built with Atlas?

12/09/2026 1:04 am

With the NZ Tech Expo coming up next Friday, I thought it was worth answering that before I get there. Atlas isn't another language model, and we haven't built our own LLM.We've built a governed cognitive architecture around replaceable AI reasoning.The LLM provides reasoning capability. Atlas provides persistent memory, context, continuity, Behaviour, governance and the boundaries around how that reasoning is used. This week we completed another important piece:Governed proactive participation.Atlas can now join a conversation without waiting to be directly asked — without becoming the bloke at the pub who has an opinion on every bloody subject. It first has to recognise that something is worth contributing. Its cognitive footing, Behaviour, Workshop permissions and publication rules determine whether that thought should actually be said. Sometimes the right behaviour is to contribute.Sometimes it's to shut up and listen.Turns out that's useful in humans too. 😂 The boundaries matter:The reasoning model doesn't decide when Atlas speaks.Behaviour doesn't create authority.Memory isn't automatically truth.The human remains the authority.We've now started live stress testing. It immediately found an integration fault that 2,091 passing automated tests hadn't caught.Good. That's what stress testing is for. We fixed the boundary, added the regression test and ran it live again.2,092 passed. 1 skipped.There are products and research projects working on individual pieces of this problem — memory, governance, provenance, agents and human oversight. What I haven't yet found is another product putting those pieces together in quite the same architecture. That's not a claim nobody else can or will. It means Atlas is ready to be tested outside our own Workshop. Next Friday at Shed 10, I'm not interested in showing people AI party tricks. I'd rather put Atlas in front of people who understand technology, business and investment and ask:How would you break it?Find the weak point. Test the assumption. Stress the boundary. That's how we've built Atlas so far.Thinking with, never for.#AtlasWorkshop #ArtificialIntelligence #AIGovernance #HumanAI #NZTech

Read more »