Now
Last updated: 2026-01-09
Current focus
Making CueCrux behave well under real-world use
Most of my time is now spent on how CueCrux behaves once ideas meet reality.
That means moving beyond conceptual clarity into operational discipline:
- observing how answers are reused, not just produced
- watching how confidence accumulates over time
- identifying where assumptions decay silently rather than fail loudly
- testing how uncertainty signals are interpreted by humans and systems
The work is less about features and more about behaviour: what the system encourages, what it discourages, and what it accidentally rewards.
Evidence, uncertainty, and confidence signalling in AI systems
The first half of 2025 was largely spent thinking, reading, and writing about AI, truth, and the nature of answers.
That phase clarified something important: the problem is not that AI systems are often wrong, but that they are confident in ways that travel.
My focus now is on practical questions that follow from that realisation:
- How should uncertainty be represented so it informs decisions rather than paralysing them?
- How does confidence inflate when answers are reused across systems and contexts?
- What signals help people rely on an answer appropriately, rather than absolutely?
These questions sit uncomfortably between technical design, human behaviour, and governance. That discomfort is where the work lives.
Writing on trust, scale, and decision fragility
I’m continuing to write, but more selectively.
The writing now functions as:
- a way to clarify assumptions before they harden into systems
- a place to examine failure modes before they appear operationally
- a record of how ideas evolve once they are tested, not just argued
The emphasis has shifted from explanation to inspection. Fewer posts, more considered ones.
Active work
Alongside the conceptual work, I’m currently focused on:
- CueCrux MVP: bringing the core ideas into a usable, inspectable form
- Release discipline: deciding what must exist at launch, and what must explicitly wait
- Founder responsibilities: structure, governance, funding paths, and early partnerships
- Boundary-setting: resisting pressure to overextend the system before its behaviour is well understood
This phase is intentionally constrained. Shipping something stable is more important than shipping something impressive.
Paused or deprioritised
Some things are deliberately on hold:
Broad AI commentary
I’m avoiding generic takes on AI progress or capability races unless they connect directly to system behaviour and trust.
Tool-level optimisation without structural impact
Faster models and clever integrations matter, but only after the underlying reasoning and signalling problems are addressed.
These may return later. They’re not useful distractions right now.
How this is likely to change
As CueCrux moves from MVP into broader use, my focus will shift again, probably towards:
- governance and accountability questions
- how systems explain themselves to non-experts
- where responsibility sits when automated answers influence real outcomes
When that happens, this page will change.
Best place to follow changes
The writing remains the best place to track how my thinking evolves.
That’s where uncertainty is worked through in public, before it shows up elsewhere.
Writing