The DXP Catalyst Update - Sept 10, 2026

Opticon 2026

EVENTS

Last week I attended a couple of Optimizely events in New York. The first was an afternoon session for media and analysts, which included hands-on time with the agent platform, a briefing on everything that would be announced the following day, and about 30 minutes with their executive team. The following day was Opticon itself at the Javits Center, with customers, prospects, partners, and a mix of other attendees. I sat in on the keynote and one of the product sessions.

The roadmap was the story this year. Optimizely spent both days repositioning itself away from being a DXP vendor and toward being an AI platform for marketing, with content management, experimentation, and personalization recast as capabilities that connect to an agent platform rather than as the products the pitch starts with. Below is what was announced and what I made of it.

Three Priorities, One Platform

Kevin Li, SVP of Product Strategy, organized the roadmap session around three priorities:

• Marketer experience (MX). End-to-end workflows and collaboration tools to get more done, faster.

• Customer experience (CX). Everything needed to launch, personalize, and convert, continuously.

• Agent experience (AX). Visibility and optimization tools to get discovered and seen by AI.

Alex Atzberger, CEO, used the same three paths in the keynote, mapped to scaling execution, driving revenue, and modernizing for AI. He described the marketer experience work as the part of the iceberg below the waterline, the campaigns, approvals, translation, compliance, and analysis that never show up in the finished output but consume most of the calendar. That resonated with the room, and it is a more grounded pitch than most agentic marketing narratives currently make.

Token Economics and Model Routing

The announcement that got the least stage time is the one with the biggest implications for how customers get billed. Optimizely introduced a family of purpose-built, post-trained models for marketing tasks, branded under Mark, alongside Mark-Bench, an open source benchmark for scoring marketing AI. Mark-Bench runs 285 tasks across 15 marketing functions and more than 6,000 criteria, covering work like writing a press release, creating a social post, and drafting email copy. The company reported early testing showing 10x greater cost efficiency than state-of-the-art general models, and published results putting its own agent platform at 67 percent all-pass against Claude Code's 60 percent at 2x lower cost.

Token economics is a real issue, and it came up during the media and analyst session. The concept I liked most here was model routing. Not every marketer or developer is up to date on every model that exists, and frontier intelligence is not required for every task. If Optimizely can apply some discretion based on the prompt and select which model best serves it, that is intriguing. We will see how it works in practice. Worth noting that the benchmark results are vendor-run and vendor-scored with no independent validation yet, though an open source benchmark other vendors can run their own models against is a more useful contribution to the category than another set of internal metrics nobody can reproduce.

Hands On With Opal

The media and analyst session included hands-on time with Opal, and I was impressed with it. Building a skill for recognizing brand tone was straightforward: feed in a few articles, web pages, or PDFs, and an assistant analyzes them and produces the tone profile. You can then use the same assistant to create an agent and have it apply that skill, all in a single prompt. Skills and context have been in the product since the beginning, but what changed is that natural language alone now builds both the skill and the agent that uses it. It minimizes even having to learn your way around the product to a degree, which lowers the barrier considerably for teams that know what good output looks like but not how to configure their way to it.

Virtual Teammates

Virtual Teammates reached general availability with five role-based coworkers at launch: Chief of Staff, SEO and AI Search Analyst, Marketing Analyst, Personalization Strategist, and CRO Manager. The design intent is persistence and proactivity. Each teammate holds organizational context over time, runs work on schedules or triggering events, and carries its own identity and permission scope rather than waiting on a prompt for every step. Multiple humans and multiple teammates can work together in a single thread.

The keynote demo followed a product marketing teammate named Paige, who surfaced a gap between analyst results and what the Optimizely site actually said, pulled Google Analytics data, diagnosed the performance issue, proposed new messaging, and generated a revised landing page, then explained the reasoning and routed the work for approval.

The proactive framing did not come out of nowhere. Back in December, Alex posted about where agents were heading in 2027 and specifically referenced ambient AI. Seeing the company ship agents that are proactive rather than reactive tracks directly with that. This is the capability I would most want to spend real time working with inside their ecosystem, because the difference between an agent that waits for a prompt and one that monitors and surfaces on its own is the difference between a tool and something closer to a colleague.

Personalization and Optimization

Limitless 1:1 Personalization is generally available, with agents generating individual landing pages per account or per buyer. Optimizely describes it as building a microsite for every target buyer, governed and continuously optimized by agents. The account-based version scales to thousands of pages; the consumer version is being pitched into large loyalty programs. AI has made it more conceivable that we could present website experiences that are genuinely one to one, but that has not been the case in practice, which is what makes this worth watching. Measurement and tracking across that volume of pages is the part I would want to see working.

Autonomous Optimization fits alongside it, entering closed beta and placing Optimizely's existing ideation, setup, creation, and analysis agents inside a continuous feedback loop, with the company reporting 2-3% conversion lifts on customer sites without human involvement. Dynamic Experiences also entered closed beta, putting an LLM into front-end composition so content is assembled at first visit and updated as interactions accumulate, with the marketer supplying input and guardrails describing what could be shown rather than building the page.

The Agent Experience Track

Agent Visibility is generally available. It takes CDN-level data and turns retrieval bot activity into reporting marketers can read, classifying which agents are hitting a site and whether they should be. Most marketing teams have no picture of what retrieval agents are doing on their own websites, and the industry spent years treating all of that traffic as something to block rather than something to understand. This capability first shipped in June alongside the Conductor partnership, as did Limitless 1:1 Personalization, so a portion of the Opticon slate was already in market before the event.

Agent-Ready Experiences, in closed beta, is Optimizely's WebMCP implementation. It exposes a structured list of actions an agent is permitted to take on a site, so a browser-resident agent can book, submit, or configure rather than screenshotting a page and guessing where to click. Nazanin, one of their VPs of Product, was talking about WebMCP a few weeks ago at CMS Connect, so it was intriguing to see it in action this quickly. The protocol is in origin trials, so timing here depends on Chrome and Edge more than it depends on Optimizely.

Final Thoughts

I gave a talk in Boston on Tuesday that touched on exactly this: vendors stepping away from the DXP label and reaching for new positioning, including terms like "agentic experience platform." Optimizely calling itself an AI platform for marketing fits that pattern. The category is renaming itself in real time, and Optimizely moved earlier and more completely than most.

The integration story is the tell on how serious the repositioning is. The agent platform is being positioned to orchestrate work across Adobe, Sitecore, Contentful and Contentstack on the content side, HubSpot for marketing automation, and tools like 6sense on the data side. That is a company describing an addressable market considerably wider than its own installed base, and it changes what an Optimizely conversation looks like for an organization already running someone else's CMS.

One last thing worth noting. Everything announced was generally available, in closed or open beta, or running live on Optimizely's own domains. Nothing was positioned as arriving two quarters from now. Vendor roadmap sessions almost never work that way, and it is a reasonable signal about how fast this part of the category is moving.