Early-stage · Prototype under active development
THOSAN ONE
One founder.
An AI workforce.
One operating system.
An AI-native operating system that discovers opportunities, coordinates execution, supports sales and continuously learns from business outcomes.
Built for solo founders and micro-businesses. The founder sets goals, budgets and approvals.
Loop status · cycle 3
- Observe7
- Discover4
- Decide3
- Build3
- Create7
- Distribute5
- Engage4
- Sell5
- Measure3
- Learn3
Claude briefing
- Two approvals are blocking the kit experiment: the $19 payment link and the Saturday community post.
- The clip-vs-carousel result is directional only; Claude recommends two more comparisons before changing the content mix.
Waiting for you
Create a payment link for the Documentation Kit at a $19 test price and publish it on the landing page
payments.create_link
Post the free checklist as a resource in Community group A on Saturday 10:00
social.publish_post
Reply to Sample Contact A confirming the kit is a one-time purchase
inbox.send_reply
Jobs
- Weekly signal scanrunning
- Publish kit price on landing pageawaiting approval
- Weekend community postawaiting approval
The problem
Most AI tools automate a task.
THOSAN ONE operates a loop.
A solo founder runs research, offers, content, distribution, customer messages, sales and reporting at the same time — across tools that don't know what the others did. The founder becomes the only integration layer, and runs out of hours.
Today: isolated tools
- A chat assistant drafts a post — without knowing what sold last month.
- A scheduler publishes it — without knowing which experiment it belongs to.
- A spreadsheet tracks leads — without knowing which content brought them.
- Results sit in an analytics tab that nobody connects back to the next decision.
THOSAN ONE: one connected loop
- Every decision is recorded with the evidence and hypothesis behind it.
- Every job, draft, post and conversation links back to that decision.
- Every outcome is measured against the experiment that caused it.
- Lessons are written to business memory and read before the next decision.
The closed loop
Decisions, execution and outcomes in one system.
THOSAN ONE remembers what happened and uses results to influence the next business decision. That feedback loop is the product.
- 01
Observe
Collect market signals, channel data and customer messages.
- 02
Discover
Turn signals into opportunities with evidence, risk and fit.
- 03
Decide
Set objectives, hypotheses and small experiments.
- 04
Build
Shape offers, pages and assets for the experiment.
- 05
Create
Concept, script and draft content tied to a campaign.
- 06
Distribute
Queue approved content for the right channels.
- 07
Engage
Classify conversations and draft replies.
- 08
Sell
Qualify leads and propose the next sales action.
- 09
Measure
Track results against each experiment's metric.
- 10
Learn
Record outcomes and lessons that shape the next decision.
Product modules
Sixteen modules, one business memory.
Each module is a view onto the same loop. The prototype runs on clearly labelled sample data.
01
Executive Dashboard
Loop status, blockers and a daily briefing.
02
Market Radar
Signals, trends, demand and competitive observations.
03
Opportunity Explorer
Scored opportunity cards with evidence and risk.
04
Strategy Planner
Objectives, hypotheses, experiments and action plans.
05
Product / Offer Builder
Offers linked to the opportunity they test.
06
Content Factory
Concepts, scripts, posts and campaign status.
07
Distribution Queue
Planned publishing with approval status.
08
Customer Inbox
Intent classification and suggested replies.
09
CRM / Lead Pipeline
Leads by stage, fit and next action.
10
Sales Agent
Lead qualification and deal status.
11
Revenue Signals
Pipeline, conversions and early revenue signals.
12
Business Memory
Decisions, outcomes, lessons and reusable context.
13
Learning Loop
Experiment → action → result → insight → next.
14
AI Workforce
Agents, active jobs, tool use and cost estimates.
15
Control Room
Projects, jobs, workers, events and audit log.
16
Human Approval Queue
Every gated action waits here for the founder.
Why Claude
Claude is the reasoning layer, not a text generator bolted on.
Running a business loop is judgment work: weighing messy evidence, planning under constraints, holding careful customer conversations and deciding what to try next. THOSAN ONE needs a model with strong reasoning, reliable tool use, structured outputs, long context for business memory and predictable behavior around instructions.
Claude = reasoning brain
Agents = specialized workforce
Tools = hands
Business Memory = organizational memory
Control Room = management & governance
Workers = execution layer
Market intelligence
Synthesizes market signals, discovers and evaluates opportunities against evidence.
Strategy & planning
Generates strategy, plans experiments and decomposes them into jobs for agents.
Coordination
Coordinates agents, selects tools and interprets structured business data.
Content intelligence
Shapes concepts and drafts around what the audience has responded to before.
Customers & sales
Classifies conversations, drafts replies, qualifies leads and assists sales.
Judgement & learning
Evaluates outputs, interprets business memory, analyzes outcomes and decides next actions.
How it is wired in the prototype
A typed reasoning endpoint calls the Claude API with structured outputs validated against schemas. Every action Claude proposes passes through a deterministic policy engine before any agent can run it. Without an API key, the endpoint returns clearly labelled demo output.
Architecture
Reasoning, execution and governance as separate layers.
Claude reasons. Agents specialize. Workers execute jobs from a queue through tool adapters. The Control Room governs what is allowed to happen, and Business Memory keeps the record.
- Event-driven: every state change is an event; every event is audited.
- Policy before execution: proposals are classified before any tool runs.
- Outcome tracking: results are linked to the decision that caused them.
Founder
Sets direction
- Goals
- Constraints
- Budgets
- Approval rules
Control Room
Management & governance
- Projects
- Jobs
- Policy engine
- Approval gates
- Audit log
Claude
Reasoning brain
- Synthesize signals
- Evaluate
- Plan & decompose
- Select tools
- Decide next action
Agents
Specialized workforce
- Market Analyst
- Opportunity Scout
- Strategist
- Content Producer
- Customer Concierge
- Sales Assistant
- +4
Tools & Workers
Hands & execution layer
- Tool adapters
- Job queue
- Workers
- Event bus
Business Memory
Organizational memory
- Decisions
- Outcomes
- Lessons
- Reusable context
Outcomes flow back up: measured results are written to Business Memory, and Claude reads them before the next decision.
Human control
Autonomous within limits the founder sets.
THOSAN ONE is not blindly autonomous. Research, drafting and analysis run on their own. Anything consequential stops at a human approval gate.
Always requires the founder's approval
- Public publishing
- Payments
- Financial actions
- Destructive actions
- Important customer-facing actions
- Changes to external systems
- Sensitive data
- Irreversible operations
- 1
Claude proposes
An action with a declared tool, summary, rationale and risk.
- 2
Policy classifies
A deterministic engine assigns risk and gates. Claude can raise risk, never waive a gate.
- 3
Founder decides
Approve, reject with a reason, or edit — from the approval queue.
- 4
Everything is logged
Proposals, decisions and executions go to an append-only audit log.
Internal dogfooding
Built inside a real, under-staffed operation.
The founder runs media and real-estate-related workflows with very few people. Those workflows are THOSAN ONE's first test environment for research, content, customer engagement, lead management and automation.
Real estate is the first test environment, not the target market. The loop — observe, decide, execute, measure, learn — is the same for any small business that sells through content and conversations.
Dogfooding means the product is shaped by daily operational pain rather than a hypothetical persona. It also means we will report what works and what does not from our own use before asking anyone else to rely on it.
Current status
- • Working web prototype with 16 modules on sample data
- • Policy engine and job state machine with unit tests
- • Claude reasoning endpoint with typed, validated outputs
- • Preparing to run on the founder's own workflows
Founder & company
THOSAN AI
THOSAN AI is a bootstrapped, early-stage software company building THOSAN ONE. It was started by founder Nguyen Viet Anh to solve an operational problem first-hand: running research, content, customer conversations and sales follow-up with too few people and too many disconnected tools.
The company is exploring how AI agents, workers, event-driven systems and human approval layers can let a small company operate more effectively — with Claude as the reasoning core and the founder in control of every consequential action.
Nguyen Viet Anh
Founder, THOSAN AI
Early access
Looking for a few solo founders to learn from.
THOSAN ONE is not available yet. If you run a small business and coordinate research, content, customers and sales mostly by yourself, tell us which workflow costs you the most time.