How we use AI in Headcore’s operations, and what we tell clients about it
On April 28, 2026, Claude joined the Headcore team. Until then, AI showed up in our work as it does at most companies: a piece of text here, an idea there. Six months later, it handles WhatsApp conversations, summarizes meetings, prepares proposals and contracts, analyzes media and produces pieces, always passing through the brand’s memory and a person’s approval before anything goes out.
This piece covers what changed, where AI still gets things wrong and the rule we now use with clients to decide what can be generated and what needs to be disclosed. It closes the first part of our series on AI agents applied to brand, which covered subagents, skills and plugins, brand memory and reviewing AI content.
Six months of AI in our operations
It came in by stages, one area per month:
| When | What started working |
|---|---|
| April 28 | Claude joins the team |
| May | An agent on WhatsApp logs the lead in the CRM and books the meeting |
| June | Meetings summarized automatically |
| July | Contracts generated and sent for signature |
| August | An internal agent takes over SEO for a client |
| September | Prospecting outreach written for 100 companies |
Each stage moved ahead only once the previous one was stable. That was on purpose: one new agent a month leaves time to see where it fails before stacking the next one on top.
What holds it up: brand memory and single-purpose agents
The change that weighed most doesn’t show on the timeline. Every brand we work with got a Brand Brain, a memory holding positioning, voice, visual rules and decisions, where every statement has a source. The agents check that memory before producing anything.
Production runs on Growth System™, six Claude Code plugins that add up to 17 subagents. Each one has a single job: one builds the Brand Brain, others produce pieces, pages and video, one measures what went live, and a review panel judges everything before delivery. We explained this design in detail in the earlier pieces of the series.
Where AI still gets it wrong
The most common mistake is answering from memory when it should check the source. The agent remembers a similar rule, or a figure it saw somewhere else, and writes with the same confidence it has when it’s right. That’s why no piece goes out without human review, and every mistake that could happen again becomes a rule in the Brand Brain.
The second one is about scale. With production moving faster, review became the bottleneck. Our answer was to separate what gets judged piece by piece, like a campaign concept, from what can be checked by sampling, like variations of the same ad.
What we tell clients
Since our first project with AI, we’ve agreed with clients that its use would be disclosed. In the Brain case for Algar Telecom, the brand symbol served as the basis for a series of AI-generated images: the same triangle made of network cable, of a circuit board and of glass.
The shape and colors stayed locked to the identity, and AI varied only the material. The client knew this from the first presentation.
With clients, we use a three-line rule:
| Line | What goes in | Examples |
|---|---|---|
| Can be generated | Variation within brand rules already approved | Supporting image based on the symbol, ad variations, text draft |
| Needs disclosure | Synthetic content someone could take for a real record | Generated photographic scene, synthetic voice, image of a person who doesn’t exist |
| Stays out | Anything that simulates proof or a real person | Testimonial, a client’s face, made-up data |
The third line is the one that protects the brand most. A fabricated testimonial or an unsourced number may go unnoticed for a while, but the trust it destroys when it surfaces is far greater than whatever was gained by using it.
What regulation already requires
In Europe, article 50 of the AI Act, in force since August 2, 2026, requires transparency for synthetic content and deepfakes: telling people when they’re talking to an AI and labeling what was generated or manipulated. For systems already on the market before that date, machine-readable labeling has a deadline of December 2, 2026.
In Brazil, Conar, Brazil’s advertising self-regulation body, published a new edition of its guide to advertising with influencers in May 2026, effective since June 1. The guide addresses AI for the first time and makes clear that the rules of the self-regulation code apply whatever technology is used, with responsibility remaining with the influencer, the advertiser and the agency.
A brand with no operations in Europe tends to feel these rules through platforms and through contracts with global companies. The compliance section of our guide to agentic AI brings together what already applies in Brazil, including the LGPD, the country’s data protection law.
What comes next
The method we put together inside our operations is going to reach beyond them. The plugins we use are becoming a tool for other teams, and Headcore Academy will teach everything from AI basics to installing and using these plugins. There’s no date yet.
In the meantime, any brand can adopt the three-line rule tomorrow: a short page in the brand book saying what can be generated, what needs disclosure and what stays out.
The full series
This piece is part of a five-part series on AI agents applied to brand work:
- Subagents and agent teams in Claude Code
- Skills and plugins in Claude Code
- CLAUDE.md and Brand Brain: brand memory for AI
- How to review AI-generated content
- How we use AI in Headcore’s operations (this piece)
Frequently asked questions
Does a brand need to disclose that it used AI in a piece of content?
It depends on the content and the market. In Europe, article 50 of the AI Act has required transparency for synthetic content and deepfakes since August 2026. In Brazil, the Conar guide says advertising rules apply whatever the technology. Under our rule, disclosure is mandatory whenever someone could take the generated content for a real record.
What is an AI use policy for a brand?
It’s a short document that says what can be generated with AI, what needs disclosure and what is off-limits altogether. It lives in the brand book or the Brand Brain and applies to the internal team, agencies and suppliers. The version we use has three lines.
Does using AI lower the value of an agency’s work?
In our experience, telling clients where AI came in didn’t take value away from the work, and it made the conversation with them easier. What the client buys is still judgment: what to do, what to approve and what not to publish.
What tasks does AI handle in Headcore’s operations?
First contact on WhatsApp with the lead logged in the CRM, meeting summaries, contract generation, SEO for one client, prospecting outreach, media analysis and the production of pieces, pages and video. Everything goes through the brand’s memory and human approval before it goes out.
What doesn’t AI do in the operation?
It doesn’t approve pieces, doesn’t decide budget and doesn’t publish without review. It also doesn’t generate testimonials, client faces or unsourced data. In the media plugin, the output doesn’t even have a field for money: budget decisions stay with a person.
Sources: AI Act, article 50 · Houthoff, on transparency and the AI Omnibus · Klalaw, on the 2026 Conar guide.
