The Twin · Agent Mesh

The Agent Mesh

The design for how AI agents work together inside our engines. Every agent reads and writes only through the Strategy Twin, and every one works to a rule a person can check.
Meet the eight agent typesHow we use AI

In short

What is the Agent Mesh?

The Agent Mesh is strategy1's design for how AI agents work together inside its engines. Eight agent types, from the Orchestrator and Research agents to the Red team and Governance agents, each have one job and one guardrail, and every agent reads and writes only through the client's Strategy Twin, never directly to the others.

Defined in the glossary: Agent MeshStrategy engineRed teaming in strategyRed-team reportEvidence ledgerAll strategy terms

Eight agent types

Each agent has one job and one guardrail.

Choose an agent to see its job and the rule it works to. Filter by the kind of guardrail.

Agent · 01 of 08

Orchestrator

Its job
Plans the run and hands work between agents
Its guardrail
Can only act through Twin actions

Reads and writes only through the Strategy Twin.

Fig 4.1Agent types arranged around the Strategy Twin, each connected to it rather than to each other.

All eight, side by side

  • Orchestrator

    Its job
    Plans the run and hands work between agents
    Its guardrail
    Can only act through Twin actions

  • Research

    Its job
    Finds and reads sources
    Its guardrail
    Every claim is stored with its source

  • Data and ontology

    Its job
    Maps your data into the Twin
    Its guardrail
    Schema changes need human approval

  • Modelling and finance

    Its job
    Builds the models behind each option
    Its guardrail
    Numbers come from deterministic code; the LLM explains, it does not calculate

  • Domain specialist

    Its job
    Applies sector knowledge
    Its guardrail
    Licensed expert sign-off where liability exists

  • Red team

    Its job
    Argues against the plan
    Its guardrail
    A “how this fails” section is required

  • Narrative

    Its job
    Drafts the memo and the board story
    Its guardrail
    Cannot add facts that are not in the evidence ledger

  • Governance and audit

    Its job
    Records approvals and checks policy
    Its guardrail
    Nothing is written to client systems without approval

Critics by design

Four critics argue against every plan.

Before an option reaches the Council, it is attacked from four sides. Two judges then check the result.

Option B · waiting for the red team

  • Critic 01

    Competitor

    How would a rival respond?

  • Critic 02

    Regulator

    What rule or permit could stop it?

  • Critic 03

    Investor

    Would the numbers survive scrutiny?

  • Critic 04

    Customer

    Would the people it depends on choose it?

Judge

Consistency judge

Does the plan hold together?

Judge

Quality judge

Does it meet the standard?

How the red-team report reaches the boardSee the engines

FAQ

Questions about AI agents in strategy work

How do AI agents work in strategy1's strategy consulting?

Inside every engine, eight agent types split the work: planning, research, data mapping, modelling, sector knowledge, red teaming, drafting and governance. Each has one job and one guardrail, all of them read and write only through the client's Strategy Twin, and people review and sign what the agents produce.

What guardrails do strategy1's AI agents follow?

Each agent works to one rule a person can check. Research stores every claim with its source, Modelling and finance takes numbers from deterministic code, Narrative cannot add facts that are not in the evidence ledger, and Governance and audit makes sure nothing is written to client systems without approval.

What does the red-team agent do?

The red-team agent argues against the plan. Before an option reaches the Strategy Council, critics attack it as a competitor, a regulator, an investor and a customer would, and two judges then check the result. A “how this fails” section is required.

Can an AI agent change a client's systems?

Not on its own. The Orchestrator can act only through Twin actions, and every change to a client system passes a policy check, the named owner's approval of a dry-run and a hash-linked audit record first.

Keep reading

Related

The Twin

The Strategy Twin

A living model of your strategic world, from six domains to governed actions.

Method

The S1 Method

Sense, Model, Simulate, Decide, Deploy, Learn. Six stages, three human gates.

Trust

Evidence and trust

Evidence ledger, confidence grades, red-team reports, expert sign-off and data protection.

Proof

Compression Index

How we measure compression: baseline time against our time, method published.

Glossary

Strategy glossary

Plain definitions of the Strategy Twin, the S1 Method, evidence ledgers, red-team reports and more.

See the agents work on your decision.

In a 72-Hour Bootcamp, the agents work on one real problem, and you see what each one produced and who approved it.

Book a 72-Hour BootcampHow we use AI