Optimizely gives marketers AI teammates with job titles and its own marketing models

Optimizely used its Opticon conference in New York this week to push its Opal AI platform from single-task agents toward something closer to staff. On August 31 the company launched Virtual Teammates, five AI personas with job descriptions, and on September 1 it followed with purpose-built marketing models and an open-source evaluation set called Mark-Bench.

The teammates come in five roles: a chief of staff that prepares meetings and briefings, an SEO and AI search analyst, a marketing analyst, a personalization strategist and a conversion-rate optimization manager. Unlike a chatbot waiting for prompts, they run on schedules, retain what they have learned about the organization, hold their own system permissions and log every action to an audit trail. Kevin Li, Optimizely’s senior vice president of product, said the point is that they “carry the work through to execution.”

The company frames the launch as an answer to tool sprawl. Its survey of more than 2,000 B2B marketing leaders found that 81% switch between two or more disconnected AI tools every week and 19% juggle four or more, while 53% said current tools capture brand facts but miss emotional resonance.

Why it matters

The second announcement is the more contentious one. Optimizely says its marketing-specific models, fed by a data layer called Mark-IQ that pulls in experimentation history and web analytics, deliver roughly ten times the cost efficiency of frontier models on marketing tasks. On Mark-Bench, which spans 285 tasks across 15 marketing functions with more than 6,000 grading criteria, the company reports its Agent Platform passed 67% of tasks against 60% for Anthropic’s Claude Code, at half the cost. The benchmark is open source, so rivals and customers can check the claim, though a vendor-built test naturally favors the vendor’s own workflow.

The move puts Optimizely, which topped Gartner’s 2025 Magic Quadrant for digital experience platforms ahead of Adobe, in direct competition with Adobe’s agent “coworkers” and Salesforce’s Agentforce. Opal reached nearly 1,700 customers and more than 172,000 agent runs by May, with annual recurring revenue up 42% quarter on quarter.

What’s next

Optimizely has not published pricing for Virtual Teammates, and its executives were candid that the routine of marketers pasting AI output between tools is on its way out. President Shafqat Islam went further, suggesting the marketing job as currently shaped may not survive intact. Whether teammates with permissions and schedules reduce headcount or simply absorb the busywork is the question CMOs will be asking as they evaluate the roadmap.