Total Information Awareness for Operators
Wire an agent into every company system so the founder sees each team unfiltered.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 60%
The mechanism replaces the management reporting chain with a direct instrument. An agent is wired into the systems where work actually happens — meetings, documents, chat, KPIs — and continuously compiles a per-person and per-team picture that the founder can query directly. Because context windows now hold far more than a person can, the summarisation that a chief of staff could never do at scale becomes routine. The critical step is the rubric: you have to state what a good employee actually looks like at your company, tune it to the culture you want, and score against that rather than a generic template. The output is decision-grade visibility — which teams are real, which are not — that survives contact with the managers who previously controlled the narrative.
Origin
Extracted from Naval. Gary Tan describes what he learned from Pedro at Brex, who runs the company with total information awareness of every team via a personal agent, and argues that other CEOs are doing the same without discussing it publicly.
Core principles
- 01Most executives are blind because they see the company through managers.
- 02A million-token context window can hold more than any human working memory.
- 03You cannot act on the bottom of an org you cannot see.
- 04What counts as a good employee has to be defined before it can be scored.
- 05Visibility is the prerequisite for a smaller, higher-leverage team.
How to run it
- 1
Connect the agent to where work happens
Wire the system into meeting recordings, documents, chat and the KPI stack so it reads the primary sources rather than curated summaries.
Pro tip Start with the systems that already retain history; retrofitting capture is the slow part.
Watch out Get legal and privacy review before ingesting employee communications — this is monitoring, and the rules differ sharply by jurisdiction.
- 2
Diarise and attribute
Split transcripts by speaker so that contributions can be attributed to people and teams rather than sitting in an undifferentiated blob.
Pro tip Attribution quality caps everything downstream; fix it before building scores.
- 3
Generate a file per person and per team
Maintain a rolling markdown file summarising what each person and team has actually shipped, said and owned. This is the artefact you query.
Pro tip Keep the raw evidence links inline so any conclusion can be traced back.
- 4
Define what good actually looks like
Write down the behaviours and characteristics that constitute a good employee at your company. This is where you tune the culture you want rather than inheriting a default.
Pro tip Include behaviours that do not show up in output metrics, such as unblocking others.
Watch out An untuned rubric will reward whoever generates the most visible text, not whoever creates the most value.
- 5
Compute scores and read them as evidence
Score against the stated rubric to surface the picture, then treat the scores as a starting point for investigation rather than as a verdict.
Pro tip Look hardest at the disagreements between the score and the manager's account — that gap is the whole point of the exercise.
Watch out The system is not perfect; anyone who acts on it as if it were will fire the wrong people with unusual confidence.
- 6
Act, with humans in the loop
Use the visibility to make the org smaller and more leveraged, but keep human judgment on every individual decision. Indiscriminate mass action destroys the culture the rubric was meant to protect.
Pro tip Fine-grained action is the advantage here; if you end up doing something blunt, you did not need the system.
Watch out Blanket reorganisations read as indiscriminate internally and leak externally as exactly that.
- 7
Keep it running
Run the compilation continuously so the picture reflects the current quarter rather than a snapshot, and revisit the rubric as the company changes.
Pro tip Tie the refresh to the planning cycle so the data is fresh when decisions are actually made.
In the wild
Gary Tan describes what he took from Pedro at Brex: a personal agent that knows the KPIs and what every direct report discusses in their meetings, giving the CEO total information awareness of each team. Tan argues this is the highest-leverage thing available to a founding team right now, and that other CEOs are already doing it but will only discuss it behind closed doors.
→ A CEO who can see each team directly rather than through the managers being evaluated.
Tan recounts sitting on boards where the honest assessment was that the company needed to cut most of its headcount to survive, but nobody could say which people, because all visibility ran through managers. The panel's answer is to tie an agent into every system, evaluate everything against a stated rubric, and produce a per-person file. Naval's caveat lands immediately: you could do it, you probably should, and almost nobody will.
→ The blocking problem shifts from a missing answer to a decision the leadership has to be willing to make.
Common mistakes
Scoring before defining good
Without an explicit, culture-tuned rubric the system optimises for legibility, promoting whoever produces the most documentation rather than the most value.
Treating the output as a verdict
The panel is explicit that it is not perfect; used as an automatic firing list it converts a visibility tool into an indiscriminate one.
Ignoring the trust cost
Comprehensive monitoring changes behaviour and can gut morale if it is discovered rather than disclosed, which is the opposite of the culture the rubric claims to protect.
Is it for you?
Best for
Founders and CEOs of organisations large enough that they no longer have direct visibility into what each team is actually doing.
Not ideal for
Small teams where the founder already works alongside everyone, or cultures and jurisdictions where this level of monitoring is unacceptable or unlawful.
From the transcript
“he has total information awareness on like every team”
“You could have like, you know, a markdown file for every single employee at Google. And you could compute a score.”
“the problem is you don't know which 80%”
From the episode
Live in the Future