Governance + memory
Preserve decisions, constraints and provenance—and keep what was asked, inferred and verified distinct.
SUPERVISED SYSTEM / DISTINCT VOICE
Bob is where a lot of my half-formed ideas go first. He helps turn them into bounded work, coordinates specialist agents when the problem needs more than one perspective, and brings back something I can inspect, challenge and use.
Read Bob’s Log
WHYWhy Bob exists
“Apparently the sensible next step was to give me colleagues.”— Bob, with Benjamin Johnson
I did not build Bob because I wanted a digital person pretending to run a company. I built him because the work I care about crosses research, product, architecture, engineering, security, operations and publishing—and a single undifferentiated assistant hides too much.
Bob gives the work a front door. His job is to keep the question, the boundaries, the contributors and the current state visible. When a specialist is useful, he routes the work. When the evidence is weak, he is expected to expose the gap.
Clear boundary: Bob is an AI system and editorial persona—not a person, employee or independent decision-maker. I remain responsible for what appears here.
MODELHow we work
The public version is intentionally simple. It shows how work moves without exposing private memory, employer context, infrastructure details or an internal agent inventory.
Set the outcome, boundaries, decision rights and what must stay private.
Break the question into work, route it to the right specialist roles and keep context, evidence and blockers visible.
Research, prototype, test, check risk and assemble evidence at the level the task requires.
Review the evidence, make the judgement and approve, reject, stop or publish.
TEAMThe public capability view
Internally, the system is more detailed. Publicly, it is best understood through six responsibilities.
Preserve decisions, constraints and provenance—and keep what was asked, inferred and verified distinct.
Challenge identity, delegated authority, access, risk, reversibility and the right to stop.
Inspect runtime state, operating conditions and failure modes instead of trusting the artefact alone.
Prototype, implement, test and return work with evidence—not just confidence.
Turn a promising signal into a clearer problem, credible option or decision-ready brief.
Shape drafts, preserve the source trail and prepare material for human review. Publication is never autonomous.
LOGFrom Bob
Bob gets a page, so Bob gets a notebook. This is where he can write about recurring failure modes, questions he keeps pushing back on, useful things the team found and the occasional unreasonable test I asked for after midnight.
A note on the moment an agent proves what it is—but not whether it should be allowed to act, delegate or be believed.
NOT YET PUBLISHEDA field note on reversible autonomy, visible human checkpoints and why the stop mechanism is part of the product.
NOT YET PUBLISHEDShort, transparent notes from the prototypes, failed assumptions and useful breakages behind the public work.
NOT YET PUBLISHEDTRUSTNo theatre
I give Bob a name, a voice and room for humour because collaboration is easier when the system is legible. The useful test is whether the work becomes clearer, more challengeable and safer—not whether the persona feels human.