Opal, Agentic AI, and the Opportunities for Marketing Workflows

We’ve all seen it, AI tools promise to make marketing faster but only a handful actually do. Opal lands closer to the second category.

After exploring the platform, attending the in-person Optimizely training and building our own agents and tools, here’s our honest review: Opal is one of the most operationally usable AI platforms we’ve seen. It solves several real problems that GPT and Claude have struggled to solve consistently, and it opens the door to attainable workflow automation.

It is not perfect. It won’t replace strategy and it requires thoughtful setup. But when implemented well, it can give marketing organizations leverage most teams haven’t experienced before.

Where Opal Stands Out

1. Instruction Fidelity Creates Unusually Clean Output

Where GPT and Claude tend to forget instructions or apply their own editorial style even when explicitly told not to, Opal adheres to client rules with near-perfect consistency.

• If you say “do not use mirrored structures like ‘It’s not just X, it’s Y,’” it follows that.
• If you train a tone for a specific executive, it uses that tone and only that tone.

That alone removes a meaningful amount of cleanup time. For teams producing high volumes of content, every hour saved compounds.

2. Agents and Workflows That Operate Like Real Systems

Opal’s agent model is not back-and-forth chat. It’s a single-shot task completion.

You trigger an agent. It executes. It hands you the output.

That constraint forces discipline and avoids the “wandering conversation” problem common in other tools. This means:

• Better repeatability
• More predictable quality
• Easier auditing of how outputs were produced

Workflows can be simple. For example, we quickly built:

• a weekly news scanner
• tone-applied draft writer
• personalization audit and recommendation

Or workflows can be expansive. For example, 

• Automated, Real-time, Data-Driven A/B Test Ideation + Launch
• Automated, Real-time Google Ads Optimizations & Creative Refresh
• A specialized agent builder

The underlying idea is straightforward: build once, reuse everywhere. Instructions and agents can be cloned across workspaces, which is a massive operational benefit for enterprise programs.

3. True Extensibility Through Tools

This is where Opal separates itself from “LLM-plus-UI” platforms.

Opal allows teams to create custom tools via API. Those tools can then be used by agents behind the scenes.

This means:

• If your marketing team has proprietary models, Opal can use them.
• If you have workflows that require external systems (CRM, CMS, analytics), Opal can orchestrate them.
• If your brand has structured frameworks, brief templates, opportunity scoring, segmentation logic, Opal can execute those directly.

For organizations operating across multiple systems and channels, this extensibility can help bring order and structure to fragmented tooling, increasing the ability to show impact and returns on AI. 

Optimizely is also building its own ecosystem in the meantime working to create a marketplace for the instructions, agents, workflows, and tools providing even more value to users. 

The Technical “Why” Behind the Quality

Opal is operating with: 

• A specialized version of Gemini 2.5 Pro
• Lower-level control over the system prompt than is publicly available
• A more blank-slate approach to base instructions
• A curated tool environment (87 tools enabled by default)
• ~10 prebuilt instructions (inactive by default) and ~20 pre-built agents (~10 active by default)

This architecture gives Opal more determinism and less model “personality,” which is exactly what enterprise teams need for brand-safe automation.

Where Opal Still Has Gaps

Opal, like any AI, is not magic. It requires realistic expectations.

1. Up-Front Setup Is Non-Negotiable

Because Opal agents are single-shot, setup matters. Poor instructions = poor outputs.Teams must invest early in:

• Tone guidelines
• Instruction sets
• Agent Prompts
• Workflow architectures
• Tool configurations
• Access governance

This is less “plug-and-play AI” and more “design-and-deploy AI.” but the payoff could be significant when implemented with proper setup.

2. Agents Don’t Replace Strategy

They accelerate execution. They enforce consistency. But they do not decide why something should exist.

Brands will still need:

• Strategists to define goals
• Analysts to interpret outputs
• Humans to evaluate edge cases

Opal removes production friction. It does not replace the underlying strategic brainpower required to run a marketing organization well.

3. Training is not Optional

Self-serve training is fine for curiosity but we do not believe it is sufficient for enterprise-grade use. Teams attempting to learn Opal passively will produce inconsistent, ungoverned outputs. A guided setup, 1:1 onboarding and/or agency services will dramatically increase adoption success.

Our Takeaway for Senior Marketing Leaders

Opal has the ability to be useful AI at an infrastructure level. When implemented well, it accelerates the work that slows teams down. The brands that invest in Opal as an operational layer will likely see the highest return.

If you want to explore whether Opal fits your team, what a pilot looks like, or how to pair Opal with your current Optimizely Experimentation or other products, we’re happy to walk through it. Reach out to samantha.mcclure@crometrics.com or send any of us a DM on LinkedIn!