The question has an owner
I set the problem, context, standards, and consequence before an agent begins. The system does not invent its own mandate.
Inside the FrankX workspace
A question becomes usefulby surviving the passes.
This worked example uses one published essay to show the source set, the specialist review, the disagreement that mattered, and the decision I made.
The linked synthesis, date, source registry, evidence grade, and limitations are public. This pass map explains the review method; it is not presented as a time-stamped execution log.
Worked example · Published July 16, 2026
Intent Architecture
Source set
DeepMind’s delegation framework, WORKBank, Ulloa et al., a QJE field study, and NIST’s active work on agent identity and authorization.
Evidence Synthesis
Mapped what delegation, workplace AI, mixed-initiative systems, and governance research support—and where the evidence is only adjacent.
Contradiction Review
Rejected the tempting claim that AI delegation gives introverts a proven advantage; current studies do not establish it.
Editorial Architecture
Turned the synthesis into a seven-field intent contract, five decision-right levels, limitations, and open questions.
Frank’s decision
Present Intent Architecture as a FrankX synthesis—not a settled academic field. Keep the evidence grade at C, name the limits, and make every consequential action cross a human boundary.
Source to publication
Four stages. One accountable chain.
The sequence is deliberately legible. A visitor should be able to see where the source ended, where an agent inferred, and where I made the decision.
- 01
Start with live material
A question, book, conversation, dataset, workshop, or unfinished system.
- 02
Run specialist passes
Agents gather sources, compare interpretations, expose gaps, and build a testable draft.
- 03
Make the human decision
Frank challenges the synthesis, cuts what is weak, and chooses what should become public.
- 04
Publish something inspectable
Research, architecture, a guide, book intelligence, a prototype, or a partnership system.
Operating principles
Scale the passes. Keep the judgment human.
The point is not to hide automation behind a personal brand. It is to make the division of labor clear enough that the output can earn trust.
Agent roles stay bounded
Research, contradiction, architecture, implementation, editorial review, and verification are separate passes with different failure conditions.
Sources survive the synthesis
A strong page lets you distinguish source material, inference, my interpretation, and the decision I made from it.
Review can stop publication
Broken evidence, generic language, privacy risk, false partnership claims, or weak production proof can stop an otherwise polished release.
What the human review asks.
Quality is not a final grammar pass. Review can change the thesis, reject the evidence, narrow the claim, or stop the release.
- Is the source real, current, and represented fairly?
- Which claims are facts, which are synthesis, and which are Frank’s judgment?
- Can a specific person use the result to decide, make, or build something?
- Does the language sound like Frank speaking to that person?
- Can the page, workflow, or product be inspected and rolled back?
Public outputs
Follow the work that matches your question.
Research
Evidence-led research hubs
Source-linked investigations across AI systems, models, creator workflows, and emerging technology.
Book intelligence
The FrankX Library
Books reconstructed as usable models, with Frank’s interpretation kept separate from the source.
Architecture
AI systems you can inspect
Reference architectures, agent workflows, evaluation boundaries, and production decisions.
Applied work
Partnership systems
Focused briefs, prototypes, and operating systems built around a person, mission, or real workflow.
Field guides
Methods from the work
Practical guides that keep the source, decision logic, and next action close to the page.
Working notes
The journal
Short dated notes from what Frank is testing, noticing, changing, and learning in public.