Enter password to view
TREVOR GILCHRIST
Gathering per-user context for optimal agentic builds
The first principle
Generic agents fail when they do not understand the executive’s real context: judgment standards, communication preferences, source boundaries, stakeholder obligations, recurring pressures, and what must never be inferred or promised.
The Context Forge method
Context Forge captures operating context through guided spoken sessions, controlled source intake, and reviewable evidence states. The process separates portable professional judgment from current-role material, overlap, hold, delete, and off-limits content.
The capture protocol
A structured Superwhisper-powered surface prompts one question at a time, captures spoken answers as segmented context, supports bookmarks, retakes, topics, asides, and continuation points, then routes material into reviewable states before it becomes durable context.
From context to agents
The platform does not start by asking which agents a person wants. It captures the expectations and demands placed on them, identifies recurring burdens and judgment gates, then designs bespoke agents only where the evidence proves they should exist.
Per-person agent portfolios
One executive may need customer expectation-risk review. Another may need board-prep synthesis, delegation-return-loop monitoring, proposal evidence-gap analysis, or style-safe communication drafting. The useful portfolio emerges from the person’s operating reality.
At enterprise scale
Individual operating assets remain private and permissioned, while approved organizational layers can compound: shared templates, review gates, role taxonomies, source-use rules, communication standards, workflow patterns, and aggregated friction signals.
The operating-context method
01
Capture
Guided spoken answers and approved source material.
02
Structure
Portable, role-bound, overlap, hold, delete, and off-limits states.
03
Infer
Repeated burdens, judgment gates, risks, and open loops.
04
Design
Bespoke agents with source rules and review gates.
05
Prove
Before/after output improvement and reduced cognitive load.
Why it matters
Executives carry high-value tacit context that generic AI cannot safely infer. Structured capture turns that context into a private operating asset, so agent design becomes evidence-led rather than speculative.
Current proof path
Build one complete executive operating asset. Use captured context to improve real outputs. Document the repeatable method before enterprise translation.
The Context Forge Question Architecture
These questions were rewritten to make the capture process feel less like an interview and more like a private calibration session for building useful executive agents.
The intent is to help the system learn how your work actually reaches you, what genuinely needs your judgment, what should stay visible without living in your head, what can be prepared before it reaches you, and what must remain private, role-bound, or off-limits.
The main standards applied were:
Concrete before abstract: Each section starts from a recent real example, not a theory of how you work.
Low threat: The wording avoids anything that sounds like evaluation, assessment, productivity review, or delegation critique.
Time respect: Questions are designed to be answerable quickly by voice, without preparation.
User control: You can answer partially, stay at pattern level, skip anything sensitive, mark material off-limits, or hold it for review.
Operational usefulness: Every answer should help create a future agent behavior, boundary, review gate, memory object, or workflow rule.
Enterprise fit: The questions are being tested not only for you, but for whether a skeptical, time-poor Nvidia executive would understand them, trust them, and answer usefully.
The central design principle was: the questions are not just intake. They are the trust surface through which the eventual system earns the right to be useful.
A clearer way to map what should stay with you
01
01/01
Think of a recent message, meeting, request, or thread that ended up back with you. What was happening, in plain terms?
01/02
What made it land with you rather than keep moving elsewhere?
01/03
What part, if any, genuinely needed your judgment?
01/04
What could someone or some system have prepared, summarized, drafted, or moved forward before it reached you?
01/05
What needed to stay visible, even if it did not need to stay in your head?
01/06
Is this example mostly about your durable way of operating, your current role, both, or something that should be kept out of the system?
01/07
If the system saw something similar later, what should it do before trying to help?
02
02/01
Think of a recent day, trip, week, or meeting cluster where the load became hard to hold. What was the situation?
02/02
What were the main things competing for attention: messages, meetings, travel, deadlines, follow-ups, decisions, people waiting on you, or something else?
02/03
Which parts genuinely mattered, and which parts were mostly loud, repetitive, or time-sensitive?
02/04
What were you forced to remember manually?
02/05
What became hard to keep visible even though it still mattered?
02/06
Where did travel, time zones, meeting density, or context switching make the situation harder?
02/07
If the system could have reduced one piece of that load without creating new work for you, what should it have handled first?
03
03/01
Think of one recent situation where something needed your judgment before it could responsibly move forward. What was happening?
03/02
What made it unsafe, premature, or inappropriate for someone else to decide alone?
03/03
What context did you need before you could answer responsibly?
03/04
What could have gone wrong if someone had answered too quickly?
03/05
What could someone else have prepared, even if the final judgment still needed to stay with you?
03/06
What signal would tell the system that this needs to rise above ordinary urgency?
03/07
If the system saw this pattern later, should it prepare, ask, hold, escalate, draft, or stop?
04
04/02
Now think of one piece of work that moved away but came back. What brought it back?
04/03
What should the system notice about the difference between those two cases?
04/04
What kinds of work need your approval but not your execution?
04/05
When someone else owns the next move, what still needs to remain visible to you?
04/06
When should something leave your field of view entirely?
04/07
If the system saw work leave your hands, what should it track, remind, hide, or ask before resurfacing?
05
05/01
What kinds of material are clearly off-limits?
05/02
What kinds of material may be discussed at pattern level but not stored?
05/03
What kinds of material may be summarized but not copied?
05/04
What kinds of material, if any, should only be used locally or temporarily?
05/05
What kinds of drafting would be useful but should always require your explicit review before use?
05/06
What actions must the system never take without asking?
05/07
Is there employer-bound material that might still produce a portable principle if abstracted and approved?
05/08
Where are you unsure enough that the safe answer is hold for review?
06
06/01
If this were useful within one week, what pressure would it reduce first?
06/02
Which first surface sounds closest: what needs judgment today, what loops are still open, what meetings need preparation, what follow-ups were created, what drafts need review, or what risks need checking?
06/03
What should be visible immediately?
06/04
What should stay hidden unless you ask for detail?
06/05
What would make the first output feel trustworthy rather than clever?
06/06
What would make you reject it immediately?
06/07
What should the system ask you before taking the next step?