Brújula Systems
Home The Model Editorial Personal ES
Brújula Personal · Compass
🚧 Under construction · expected launch Q4 2026

An analyst, not a dashboard

A personal agent that shares the same world model as the newsroom — with persistent memory of your context, your decisions, and your bets. Not shipping yet, but you can join the waitlist.

Join the waitlist → See a sample letter → See the model →
Agent · diagnostic mode
30% of my portfolio is in semiconductors. What does the model say this week?
TM-4 (physical bottleneck) rose +2 this week: 56%. Two anchors — datacenter energy and advanced-chip concentration. For your exposure, the risk is not demand — it is the electricity constraint. Logged on June 24 in your journal; still active.
TM-4 · 56% Anchor · Federal Register

Three modes, one model

The agent works on the same thesis-and-probability structure you see in public — but applies it to your situation.

Mode 01

Diagnostic

Where you stand vs the model. The agent maps your context against active theses and tells you which forces affect you and by how much.

Mode 02

Exploratory

What if. Walk through alternative scenarios on the same probabilities, without leaving the audit trail to sources.

Mode 03

Generative

What you'd do about it. Proposes moves and puts them through defensive and constructive stress tests — always as an instrument, never a recommendation.

Dynamic journal

Your situation, updated as the world moves

Not a chat log. A living record of your context that the agent revisits every time the model moves. When a thesis that affects you changes, you know — with the anchor, the magnitude, and the date.

Jun 24
Logged semiconductor exposure (~30% portfolio). Link to TM-4.
TM-4 · physical bottleneck
Jun 27
TM-4 up +2. The constraint is electricity, not demand. Review before Q3.
▲ +2 · Federal Register
Jun 30
Institutional trust (TM-7) at 95%. Possible effect on regulatory horizon.
TM-7 · 95%

Personal access

Compass
Coming soon
·Personal agent with persistent memory (Q4 2026)
·Synthesis adapted to your context and decisions (Q4 2026)
·Three modes: diagnostic, exploratory, generative (Q4 2026)
·Dynamic journal, updated as things change (Q4 2026)
Join waitlist →
Institutional
Terms on request
·Private theses — isolate proprietary hypotheses
·Correlation with the public model, auditable
·Decision stress testing under scenarios
·Dedicated editorial desk · API access
Contact →
📄 Fictional profile · demonstration

A Compass letter for Mateo

Mateo is a fictional archetype: senior software engineer, 30-39, based in Bogotá, remote for a Texas company, paid in USD and living in COP. He's weighing whether to specialize in AI, change roles, or emigrate over the next 12 months. Any resemblance to a specific real subscriber is coincidental; the profile exists to show what Compass writes when you tell it who you are and what you're deciding.

The essentials. This week the board has two stories, and only one touches you directly. The active war between the United States and Iran — ten rounds of strikes in seven days — pushed three theses to elevated levels: Taiwan Strait conflict, fiscal crisis in advanced economies, nuclear-weapon-use risk. That vector dominates the rest of the system; for your situation it matters as context, not as decision. The acceleration in knowledge-work automation, by contrast, crossed structural thresholds on three separate fronts, and its dynamic is the opposite of the war: it moves in years and doesn't answer to a ceasefire.

Autonomous agentic capability before 2027 (0.67 → 0.73). Three convergent signals in one week: Kimi K3 reaches the state of the art on agentic benchmarks, ANet Patu-1 shows emergent coordination in heterogeneous agent networks, and AutoSynthesis runs end-to-end scientific meta-analyses without constant human intervention. AutoSynthesis matters to you specifically: it chains steps across a weeks-long project in a specialized domain, which is exactly what the thesis measures. It's now the AI thesis with the highest probability in the system.

White-collar labor disruption before 2028 (0.522 → 0.637). Three events with distinct mechanisms that reinforce each other: 4,500+ Google workers signed a petition to the CEO asking for AI-layoff protections; 26 Meta employees filed a federal lawsuit alleging the company used AI to select layoff targets; Nine Entertainment cut 30 positions from its flagship papers explicitly citing AI's "extreme" disruption as the cause. When the thesis crosses 0.60, the historical window for specialization-based differentiation in the sector narrows; for you, that's a signal to review whether your current profile still sits inside that window, not a verdict on your individual situation.

The favorable read — the application layer isn't saturated. Two of this week's moves point the same way: sustained price drops in second-tier models (58% on the most recent round) and the acceleration of agentic systems in specialized domains. What that opens is space in the application layer — building with those models for a concrete domain — that isn't saturated yet. Frontier models concentrate in three or four actors; the advantage of applying those models well to a real work domain is much more distributed. That's the favorable force in your week.

For your decisions. Over the next 3–4 weeks, publish a working demonstration of an agentic project applied to your current domain — a verifiable artifact, not a résumé. That object works for you in two directions at once: it positions you in the application layer, and it gives you real data on how your profile is received in the current market, without locking in the specialization, role-change, or migration decision yet. It's information before commitment, and it's entirely under your control.

Closing. The board this week runs two stories that don't merge: the Gulf war, which dominates geopolitical motion and is largely reversible in weeks; and the acceleration in knowledge-work automation, which moves in years. Over a 12-month horizon, the second is the one that describes the terrain where your career decision is played, with the important caveat that the application-layer space is real, not a consolation. The board describes the terrain; the decision on when and how you move remains yours, and depends on variables — your network, your concrete economic conditions, your real risk tolerance — that this analysis doesn't measure.

This is orientation, not a prediction about Mateo's life. Compass is in design; to receive this kind of correspondence when the product is ready, join the waitlist at no cost.

Join the waitlist →

Join the Compass waitlist

Compass is still under construction. When it ships (expected Q4 2026) you'll hear from us first. No charge, no trial, no calendar hold. Role and horizon help us segment the launch — required for now.

We only store email, role and horizon · no marketing
By joining the waitlist you accept the Terms and Privacy Policy.
Already a customer? Manage your subscription →
© 2026 · Building in public Privacy Terms
Read today's editorial →