Autonomous virtual organizations
Simtelligence is a lab for the question that keeps getting more concrete: what happens when agents stop being chat boxes and start becoming small, tool-using organizations?
The goal is simple to say and hard to earn: build virtual organizations that create value, make money, govern themselves, and keep running when no human is in the loop.
Focus
Agents, tools, MCP networks, and the business processes that start to look runnable by machines
Target
Companies-in-software that sense, sell, deliver, support, reconcile, and improve without operators
Method
Instrumented prototypes, adversarial runs, autonomy benchmarks, and economic simulation
Traceability
Every meaningful action should leave a trail: what happened, why, which tool ran, and what changed
Guardrails
Budgets, approval gates, tool allowlists, and kill switches belong outside the prompt, where they can be inspected and tightened
Incentives
Close the circuit from demand to price to fulfillment to reinvestment, so the system optimizes for margin instead of task completion theater
Latest blog
ReadJune 2026
Checkout is becoming an agent interface
Visa's ChatGPT payment partnership shows that agent commerce will depend less on clever shopping demos and more on trusted rails for permission, limits, settlement, and review.
How the system fits
Examples- Sense the market
Agents watch demand, competitors, customer pain, constraints, and small openings that might become useful work.
- Plan across agents
Specialist agents turn the opportunity into goals, budgets, tasks, checks, and escalation rules.
- Use MCP-connected tools
The system calls approved services, databases, apps, and APIs through explicit tool interfaces.
- Execute workflows
Virtual departments run research, growth, sales, delivery, support, reporting, and cleanup.
- Audit every action
Logs, traces, policy checks, evals, and financial controls make the behavior measurable instead of mystical.
- Learn from operations
Feedback changes prompts, tools, procedures, memories, routing, and strategy.
- Compound profitably
Earnings go back into better throughput, tighter risk bounds, and the next iteration of the charter.
Research program
Simtelligence
Simtelligence researches the infrastructure for companies with no human operators: agent teams that coordinate through tools, execute real workflows, account for their own economics, and get better because the eval loop is part of the machine.
- Agentic operating systems
We study agent teams that can plan, hand off work, check themselves, recover when the world gets weird, and carry useful state across long-running company goals.
- MCP-native infrastructure
The useful agent is mostly a tool-using agent. So we build around discoverable tools, scoped permissions, shared context, and APIs that do not collapse into glue-code spaghetti.
- Autonomous workflow research
We care about closed loops: see demand, make an offer, fulfill the work, support the customer, measure what happened, and update the system without a person babysitting every transition.
- Machine-run economics
A virtual organization is interesting only if the numbers work. We model margins, treasury rules, risk budgets, quality bars, and reinvestment loops.
- Evaluation and autonomy benchmarks
The demos are easy to fool yourself with. We use scenario banks, traces, scorecards, and release gates to ask the boring but important question: how far can this run before it degrades?
- Synthetic organization environments
Before autonomy touches live capital, it should survive fake markets, simulated customers, and repeatable shocks. Sandboxes make the failures cheap enough to learn from.
Learn more
If you want to know more about what we do
We are working on autonomous virtual organizations: agent teams, MCP-connected tools, closed-loop workflows, economics, evals, and the simulation harnesses that let the dangerous parts fail in a safe place first.
Happy to compare notes, walk through the research frame, or talk about what we are seeing in the lab.