I have spent this year covering the creative and design platforms alongside the enterprise AI infrastructure work, and for most of it the two felt like separate beats. Adobe in one week, a contact center architecture review the next. Different rooms, different vocabulary, different buyers. That stopped being true somewhere around the middle of this year, and I want to lay out why.
Three strategies that look nothing alike
Watch Adobe, Figma and Canva across 2026 and you would swear they were in different businesses.
Adobe spent Summit in April retiring the Experience Cloud umbrella in favor of CX Enterprise, organized around brand visibility, customer engagement and the content supply chain. The headline was Firefly AI Assistant, which takes a described outcome and orchestrates the steps across Firefly, Photoshop, Premiere, Lightroom, Illustrator and Express. By June the creative agent had reached InDesign and Frame.io, and Adobe had started publishing its creative tools into ChatGPT, Claude, Copilot, Gemini and Slack.
Figma spent Config in June arguing that code belongs on the canvas. Code layers pull repositories and code flows into the file and turn design layers into interactive code. Motion and shaders landed on the same surface. Dylan Field put it as “Code is material for design.” The AI agent learned to build reusable skills from a text prompt and to connect out to Notion, GitHub and Excel.
Canva spent Create in April shipping more than forty features, led by Canva AI 2.0, which it describes as a foundation model built for creativity. Brand System moved inside Affinity. Cavalry went free. Affinity picked up integrations with Capture One, DaVinci Resolve and Claude.
Adobe is selling an enterprise content supply chain. Figma is selling the product team a shared surface. Canva is selling a creative operating system that starts free and grows into the business. Different buyers, different price points, different arguments. Three companies telling three genuinely distinct stories about where creative work is going.
One architecture underneath
Now stop looking at what each one is selling and look at what each one had to build in order to sell it.
Every one of them needed somewhere to hold brand context, so that generated output arrives on brand rather than merely plausible. Adobe calls that Design Intelligence and the content supply chain. Canva calls it Brand System. Figma calls it the design system, which it has called it for years, and which turns out to have been the right primitive all along.
Every one of them needed provenance. Adobe co-founded C2PA in 2021 and has pushed Content Credentials into the enterprise. The coalition passed six thousand members in January. Samsung and Google now sign at the point of capture. OpenAI, Google, Meta and Amazon embed credentials in generated output. LinkedIn, TikTok and Cloudflare preserve them in transit.
Every one of them concluded it could not own the model. Adobe routes to third party models and now publishes into rival assistants. Figma is model agnostic by design. Canva ships Affinity integrations with Claude. Three companies with every commercial reason to lock the model layer decided independently that they could not.
And every one of them had to reach outside its own boundary for context. Figma’s agent connects to Notion, GitHub and Excel. Adobe’s agent runs across six of its own applications and then out into other vendors’ assistants. Canva’s professional suite talks to Capture One and DaVinci Resolve.
Brand context that persists across tools. Provenance you can audit. Model choice that is not locked. Context that survives crossing a vendor boundary. That is not four independent product decisions. That is a universal context layer, and the creative tools market built one without ever needing the phrase.
I have been making this argument about agentic AI in the enterprise, most recently in CIO in April. What I did not expect was to find the same architecture arriving from the design side at the same time, from three companies competing directly for the same budgets. That matters more than it might look. Convergence between partners tells you very little. Convergence between rivals who have every commercial reason to differentiate tells you they each hit the same constraint and could not engineer their way around it.
Why it showed up here first
This is the part I find genuinely interesting. Creative work has always been the most context dependent work in the enterprise, and nobody described it that way because the context was maintained by people.
A brand guideline is a context layer written for humans. A design system is a context layer with version control on it. Creative teams have spent a decade building and maintaining structured, machine adjacent context as a matter of routine professional practice. So when generative models arrived, the creative tools already had somewhere to put the context. Most enterprise software did not, which is why so much of it is still connecting systems and calling that readiness.
If you want to know what agentic architecture looks like when the context problem was solved before the agents showed up, this is the market to watch.
The deadline that did not move
Here is where the timing stops being academic.
The EU’s Digital Omnibus on AI entered into force on 27 July 2026 and deferred the AI Act’s high risk obligations to 2 December 2027. That got reported broadly as the AI Act being delayed. For synthetic media it was not.
Article 50, which carries the transparency duties for chatbots, synthetic media generators and deepfakes, was left out of that deferral. It applies from 2 August 2026 and national market surveillance authorities can enforce from that date. The one carve out is narrow. The machine readable marking duty in Article 50(2) has until 2 December 2026, and only for systems already on the market before August. California is moving on a parallel track, with SB 942 effective in January 2026 and AB 853 following in January 2027.
Which means provenance is not a differentiator. From 2 August 2026 it is a legal requirement in the EU for exactly the category these three sell into, and the marking obligation closes on the rest of the market on 2 December 2026. Adobe co-founded C2PA in 2021, which now looks less like corporate positioning and more like reading the room five years early. The provenance question I raised about Figma and Canva last year is no longer a positioning question. It is a compliance question, and their enterprise buyers are the ones who carry the exposure.
What I would ask a vendor now
If you are buying any of this for an organization rather than for yourself, the demo is not the thing to interrogate. Four questions are.
Where does brand context actually live, and does it survive when work moves between your tools and the ones you do not own? Can you produce, on request, a machine readable record of what was generated, by which model, from which inputs? If we replace the underlying model next year, what breaks and how long does it take? And can we scope what an agent is allowed to see, per task, rather than granting it a standing role across the workspace?
Those are the same four questions I would ask a contact center platform or a data infrastructure vendor. That is the point. The creative stack is not a special case. It is just further along.
The wider read
The reason this matters beyond design tools is that the creative platforms are functioning as a leading indicator. They hit the context problem first because their users had already been solving it manually for years. Whatever architecture settles here is going to look a great deal like the pattern the rest of the enterprise adopts over the next two years, and the vendors who get there late will be the ones still arguing about which model is smartest.
The friction never sat in the model. It sat in everything the model could not see. That is as true of a design file as it is of a customer record.
I write about this as Continuous Flow Architecture, and the piece of it under discussion here is the Universal Context Layer.