HALO Solo.
A single person can build and operate an entire company. The human bottleneck is inevitable. The methodology must protect it, not saturate it. This is the operating system for building a real business with AI agents at your side.
Seven principles. Non-negotiable.
A one-person company is not a small company. It is a normal company with one human and many agents. Methodology protects the human; tooling extends them.
The unit of work is no longer the person-hour. It is the capability-outcome: who or what can deliver this, and what signal proves it landed.
Agents come in two kinds. Assistants work with you in the room. Workers run alone, on webhooks, while you sleep. You design both. You own both.
Output is cheap. Outcome is rare. Mark Done only what has produced a real signal — a paid invoice, a closed ticket, a tracked conversion. Anything else is theatre.
Every domain has a boundary. There is what agents can do, and there is what only the human can do. Confusing the two breaks the company.
Verification is not optional. You are judge and jury. The methodology forces you to play both roles in different sessions, with different evidence, on different days.
Every autonomous worker has a kill-switch. If you can't turn it off in under 60 seconds, you don't own it — it owns you.
The real failure modes of a one-person company.
A solopreneur with AI agents looks like a small army. But without method, that army works against you, not for you. These are the four failure modes that kill one-person companies before they reach traction.
Output without outcome
Agents let you produce 10× more. You start validating 1× less. Soon you're shipping things nobody asked for, faster than ever.
The bottleneck pretending to be a team
You keep saying "we" but every decision still funnels through one human — you. When that human is tired, sick, or thinking, the whole company stops.
Autonomous workers, no oversight
Agents handle support, send outreach, close tickets — and one bad prompt sends 400 wrong emails before you wake up. No kill-switch, no logs, no recovery.
Project hopping at machine speed
Agents make every new project feel cheap to start. You end up with five half-built businesses, none generating signal, all draining attention.
Two kinds of agents. Two different relationships.
HALO Solo introduces a fundamental distinction. Most solopreneurs only use the first kind. The leverage is in the second — but it changes what the methodology has to protect.
Assistant agents
They work with you in the room. You give them context, they produce, you review. Synchronous. Bounded by your attention.
- Code generation and refactor
- Drafting, copy, research synthesis
- Data analysis on demand
- Spec writing, documentation
Worker agents
They run on webhooks, schedules, queues. They respond to events while you sleep. Asynchronous. Bounded by playbook and kill-switch.
- Inbound support and ticket triage
- Lead qualification and outreach
- Recurring content generation
- Monitoring, alerting, first-response
What only you do. What agents may do.
In a one-person company with agents, the methodology fails when this line blurs. The discipline is to keep it visible at all times.
- Strategic pricing and positioning
- Firing or accepting a client
- Pivoting or killing a product
- Responding to high-value upset customers
- Signing any contract above threshold
- Final hiring or partnership decisions
- Any irreversible legal or financial action
- Public statements that bind the brand
- First-line support, FAQs, ticket triage
- Lead qualification and CRM updates
- Outbound sequences within templates
- Content drafting and distribution
- Recurring reports and dashboards
- Routine bookkeeping and data entry
- Monitoring, anomaly detection, paging
- Code, tests, refactors with review gates
Seven domains. One human across all of them.
Your one-person company has the same domains as any company. The job of the methodology is to define what agents own and where the boundary sits.
Tablero 2D, three rotating roles, Signal Wait. Building is where assistant agents shine and worker agents are dangerous.
Cold outreach is automatable; relationship is not. Handoff to human at first reply with intent.
Tiered support: Workers handle N1, auto-escalate to human on negative sentiment or churn signals.
Human writes the angle; Workers handle adaptation, scheduling, distribution and SEO meta.
Agents are perfect here: billing, dunning, VAT drafts, document filing. Human signs everything.
Workers do continuous listening so the human can do periodic deciding. Aggregate analytics and classify feedback.
The meta-domain. Watches the other six. Monitors agent behavior and inference spend.
Three hats. Never two at the same time.
The discipline of switching hats is what replaces the team you don't have. With workers running 24/7, the discipline matters more.
Loop Owner
You decide what enters the system. You review signals and worker agent logs. You decide which projects NOT to touch.
Builder
You execute with assistant agents. No meta-decisions. New ideas go to Intake. Workers run in the background.
Verifier & Signal Reader
The most critical hat. Audit workers, read production signals, mark Done what has signal. Kill what won't.
A three-column board. Build, run, watch.
We add a third execution column — workers run on their own. Visualize what assistants are helping with and what workers are doing without you.
Five ceremonies. Workers attend none.
v2 adds a weekly worker audit — because autonomous agents need explicit human review or they drift.
Loop Sync
Daily Check
Worker Audit
Signal Read
Capability & Cost Review
Six numbers. Forget the rest.
As a one-person company, six metrics matter. The first four are personal discipline. The last two govern your autonomous workforce.
Signal-to-Ship Ratio
Project Focus Index
Verifier Honesty Score
Cost per Outcome
Worker Drift Rate
Boundary Breach Count
Every worker has an off. You own all of them.
The cost of an unsupervised worker going wrong scales with its reach. Defined kill-switch for every autonomous worker.
Hard kill
Worker is producing wrong or unauthorized output.
Rate limit
Worker running at suspicious volume.
Spend cap
Worker consuming budget beyond ceiling.
Escalation trigger
Worker hits case it shouldn't handle (anger, legal).
Drift detection
Output diverges from baseline tone or accuracy.
Failure modes HALO Solo cures.
Iterating 4 hours with an assistant on something you'd have done in 90 min.
A worker is running but you can't remember exactly what it does.
The agent handled the angry customer once, so you let it continue.
Opening another project every time one gets hard.
You mark Done everything because you're judge and jury.
Paying for 14 SaaS tools without knowing which one earns its keep.
A worker running for weeks without any human audit.
You don't need everything. Start with one domain.
v2 is comprehensive — build it domain by domain. Pick the one leaking the most time.
Map your seven domains and pick the critical one.
Draw the boundary: what only you do, what agents may do.
Ship your first worker with kill-switch and audit.