Havas Doesn't Pretend

My favourite film about AI is Ex Machina. Brilliant performances, taut direction, and it made me think harder about artificial intelligence than any whitepaper ever has.

The AI in the film is named Ava. She's charming, sophisticated, and ultimately manipulative. Things don't end well for the humans who thought they controlled her.

I'm guessing CEO Yannick Bolloré hasn't seen the film. Because AVA is the the name of their new AI platform rolled out at CES. The press release explains the name comes "from the heart of Havas itself”, the letters in the middle of the company name…. Still…

What's interesting about the announcement is that while WPP, Publicis, and Omnicom race to convince you they're building proprietary AI, Havas admitted something their larger competitors won't.

They're not.

That honesty might be the smartest strategy in the room. Or it might just be what you say when you're too small to afford the pretense.

Last week, I wrote about WPP's "Super Agents”, tools their own CTO called "almost agentic" while the press release claimed breakthrough autonomous capabilities. The gap between marketing language and technical reality is now the industry norm.

Because there are only three viable paths for agencies to harness AI:

This is what Publicis bet on with CoreAI. What Accenture Song is pursuing with genuinely proprietary development. Multi-year R&D investment. Real technical talent. Actual model training on proprietary data.

It's expensive. It's risky. And if you pull it off, it's differentiated and can change the revenue model to SaaS in the world of output-based pricing.

Publicis committed to this path years ago. They have the Epsilon data infrastructure to justify it, one of the world's largest collections of consumer transaction data. The technical team from the earlier Sapient acquisition to execute it with the Bodhi platform. And the debt-free status, cash flow and probably most importantly the patience to see it through (so much so, that they alone did not need to be on stage at CES talking about it)

This isn't just buying API access and wrapping it in branded UI. It's building foundational capabilities that competitors can't easily replicate. Training models on proprietary data sets. Creating genuinely new intellectual property.

Build requires scale, conviction, and the financial runway to absorb years of R&D before seeing return.

That's why so few are genuinely pursuing this path, despite how many claim to be.

This is what Pencil does brilliantly. They're not building AI models. They're building sophisticated workflow orchestration around existing models. Making commercial AI accessible through product thinking and UI onramps for clients, not model training.

The distinction matters. Pencil are competing on understanding creative workflows well enough to make commercial AI genuinely useful for specific marketing tasks.

Speed to market. Lower R&D cost. Focus on user experience and implementation rather than fundamental research.

And critically: honesty about what they are. Pencil doesn't claim to have built proprietary AI. They built better ways to use AI that already exists. That clarity lets clients evaluate them on the right criteria: workflow sophistication, integration quality, implementation support.

Focus requires product sophistication, deep domain expertise, and the courage to admit you're not competing on foundational technology. When done right, it's a legitimate strategy that delivers real value.

When done wrong, it's just Path 3 with slightly more honest marketing.

This is repackaging existing capabilities with AI branding and hoping clients don't look under the hood, or slapping a coat of brand paint on off-the-shelf tools of others.

WPP's "Agent Hub" features "Super Agents" with "150 years of creative intelligence" and "30 years of Brand Asset Valuator data." Which sounds impressive until you realise they're describing database access through conversational interfaces.

Their CTO Stephan Pretorius, when describing these tools to Campaign, called them a move "from things that do tasks to almost agentic products.".

Omnicom's Omni+ announcement positioned integration of newly-acquired IPG assets such as Acxiom's data infrastructure, Flywheel Commerce Cloud and Interact as AI breakthrough. But integrating data systems you now own through acquisition isn't the same as building new AI capability. It's infrastructure work dressed up as innovation.

As I write this now, I think that Omnicom may actually be more concerning, it's like they don't know they don't know, while I think the WPP knows better, really.

The pattern is everywhere. Announce platforms. Demonstrate capabilities that already existed. Claim transformation. Hope clients don't realise they could have accessed most of this functionality through direct API contracts.

When every vendor overclaims, CMOs lose the ability to distinguish genuine capability from sophisticated theater. And the entire industry's credibility erodes.

Pretend requires credulous press, clients who don't ask technical questions, and enough budget to maintain the illusion until the next reorg provides cover to quietly sunset the overpromised capability.

AVA doesn't pretend to be Path 1. CEO Yannick Bolloré was explicit at CES: "AVA securely unifies leading AI in one portal for our teams and clients." Not "our AI." Leading AI, meaning GPT-5 from OpenAI, Claude Opus 4.5 from Anthropic, Gemini 3 from Google.

From the announcement: AVA is positioned as a "compliance-minded gateway" providing "secure, centralised access" to commercial models. The technical positioning calls it a "global LLM portal" that allows teams to "choose which AI model for specific tasks."

Translation: Havas is Path 2. They're building orchestration, governance, and workflow around models built by others.

And unlike WPP and Omnicom, they're admitting it.

The announcement includes specific commitments: €400 million investment through 2027, mandatory AI literacy training for all 23,000 employees as "requirement rather than specialisation," partnership with Akkio for "agentic capabilities," and spring 2026 rollout timeline.

At CES, Bolloré demonstrated AVA's capabilities by bringing an audience volunteer on stage and generating an AI-powered movie trailer starring that person. Flashy demo, but notably using Havas's existing Vermeer platform (part of their content-at-scale network) rather than showing AVA's actual governance and orchestration capabilities.

Which brings us to the strategic question: Is Havas's multi-model approach genuine Path 2 focus, or Path 3 pretense with more honest initial positioning?

The cynical read: Havas is the smallest major holding company. At €2.5 billion valuation, they're dwarfed by Omnicom's $30 billion annual revenue and WPP's scale. They can't afford Path 1's multi-year R&D investment. They can't afford Path 3's marketing theatre when clients eventually ask to see the actual capabilities.

So they're making virtue of necessity. Positioning orchestration as strategy rather than admitting limitation.

If the multi-model future is already here, and their data suggests it is, then betting on orchestration might beat betting on proprietary models that get obsoleted by the next GPT or Claude release.

eMarketer's analysis of the CES launches noted: "Havas is making a slightly different bet: that agencies will increasingly live in a multi-model world, and the competitive advantage will come from controlling how models are used, securely, consistently and in a way clients will trust."

That's legitimately contrarian positioning. While competitors rush to claim proprietary AI capabilities, Havas is betting the real value is governance and orchestration of best-in-class commercial models.

The biggest risk for Path 3 (Pretend) isn't that it fails technically. It's that clients get tired of the overclaim.

When WPP calls knowledge management tools "Super Agents" and their own CTO won't fully commit to the label, they're training clients to distrust vendor language. When Omnicom announces asset integration as AI breakthrough, they're making every future AI claim suspect.

That erosion of trust accelerates when clients start asking the technical questions that expose the gap. "Can your agent coordinate across multiple systems to achieve a goal independently?" leads to uncomfortable conversations about what "agent" actually means in the implementation.

Havas's approach, admitting you're orchestrating rather than building, might look like weakness. But it could be the first honest conversation CMOs have had about agency AI capabilities in three years of breathless announcements.

And in a market where trust is eroding faster than capabilities are advancing, honesty might be the only differentiation that matters.

Because the future of agency-client relationships won't be determined by which AI strategy succeeds. It'll be determined by which agencies stop pretending and start delivering.

Havas might have just made that choice. Whether through strategic clarity or financial constraint, they're at least being honest about what they're building.

That's more than you can say for most of their competitors.

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 22 January 2026 in the Brandflow newsletter on LinkedIn.