An agency adopts its own agent roster to run paid media at Turm

Turm Agencia manages digital advertising for its own portfolio of clients. Every account has its budget, its audiences, its credentials and its confidentiality.

For an agency, automating campaign operations brings a risk a single company does not have: the mistake does not stay inside, it crosses from one client to another.

And manual execution, which is safe because it is slow, has a ceiling.

• Scale limited by available hours: research, copywriting, setup and optimization consume the person who should be thinking strategy.
• Confidentiality across accounts: advertising platforms usually give one credential access to many accounts, and choosing by name is choosing at random.
• Advertising compliance: every platform has rules, and breaking them costs the account.
• Someone else budget: a spending mistake is not your own money, it is a client money.

A report containing another company analytics looks exactly like a correct one.

We implemented the H6 Orchestra multi-agent framework inside the agency, building a roster with bounded roles, interface contracts and per client isolation that acts before every call to the platforms.

A roster specialized by capability

• A research and competitive analysis agent.
• A copywriting agent for ad text and persuasive variants.
• A creative adaptation and visual assets agent.
• A tactical setup and audience segmentation agent.
• An algorithmic optimization and continuous monitoring agent.
• Each with bounded responsibilities, because a role that does everything cannot be audited.

Governance through interface contracts

• Formal communication protocols between agents using structured data schemas.
• Automatic QA gates and mandatory business rules.
• Validation of advertising compliance, brand consistency and budget limits happens before any call to the advertising APIs, not after the money is spent.

Strict per client isolation

• Logical partitioning and full segregation of data, tokens, credentials and ad accounts.
• One client per context, so cross contamination is not available as a mistake.
• For an agency this is not a technical improvement: it is the condition for being able to use agents at all.

Omnichannel activation and experimentation

• Automated API flows for bulk campaign creation and programmatic ad deployment.
• Standardized parameterization, which is what makes one account comparable to another.
• A continuous A/B testing engine across copy, creatives and audiences.
• Budget reallocation aimed at maximizing return on ad spend and bringing cost per acquisition to target thresholds.

Observability and the human brake

• Traceability of the decisions taken by each agent.
• Automated performance reporting.
• Routing to human approval on spending anomalies or critical changes: the system runs on its own except when it should not.

How generative AI is integrated and governed

• The solution integrates generative artificial intelligence services: language models that write, converse, extract and structure inside the workflow, rather than a rules engine with prewritten answers.
• Consumption is measured and billed in tokens, separating input, output and cache write and read, which are different line items with different prices and are accounted for separately.
• Telemetry attributes consumption per agent and per client, so the token budget is bounded and controlled per project instead of being read as a single invoice.
• Limits live in the record, not in the model: a record declares roles and caps, and which provider each role maps to is configuration that changes without touching the agent.
• The architecture is multi provider: the provider is fixed per agent and per session, never per request, and automatic retry against a different provider is deliberately switched off. If the provider fails, the agent stops and says so instead of silently continuing on another one.
• Context caching is the main cost lever, because rereading context accounts for most of the token volume in any agent operation. Sustaining the cache is what keeps the same task from costing several times more.
• And the minimum context window is declared per agent: a provider that does not cover it is ruled out by architecture, not by price.

Learn more about AI agents.

The agency stopped having an AI tool and started having its own team of agents, governed by the same rules it would apply to a human team.

What is running

• A roster of 5 roles specialized in research, copywriting, creative, setup and optimization, across 2 advertising platforms.
• Interface contracts with structured schemas and QA gates before every call to the platforms.
• Multi tenant isolation with segregated data, tokens, credentials and accounts per client.
• Programmatic activation on the advertising platforms with standardized parameterization.
• Continuous experimentation and dynamic budget reallocation.
• Decision traceability and human routing on spending anomalies.

What changes in the agency business

• The ceiling on managed accounts stops being set by the team hours.
• The team moves away from repetitive execution and toward strategy and the client relationship.
• And confidentiality across accounts goes from a promise to a control that stops the work.

Why it matters

• It is real third party adoption, not an internal demo: an outside company brought the framework into its productive operation.
• And it proves the hardest isolation case, which is that of whoever works for many clients at once.