From Agentic Theory to Practicality

Back in my Agentic Future post, I explored where Optimizely’s AI tooling might be heading - toward more autonomous, purpose-driven agents that behave more like collaborators than glorified autocomplete engines.

Since then, we’ve started to see that theory become reality in small but meaningful ways. One feature in particular stands out: Instructions, now available in Optimizely Opal, which has launched as a standalone product within the Optimizely One platform.

At first glance, Instructions might look like just another content tagging mechanism - but in practice, they lay the groundwork for true agentic behaviour in content production.

Recap: What is agentic AI?

To quickly recap: "agentic" AI systems don’t just respond to input; they operate with memory, purpose, and a degree of autonomy. Rather than micromanaging each prompt, you define behaviours and constraints, and the AI adapts to the context.

Opal’s Instructions move us closer to that vision - especially for content teams balancing multiple brands, tones, and markets.

Introducing Opal’s Instructions

Instructions in Opal are predefined, reusable prompt behaviours that guide AI-powered content creation. Rather than repeatedly explaining tone, structure, or purpose, you create Instructions once and reuse them across campaigns.

Each Instruction includes:

• A name (e.g. “Corporate Tone – Cautious & Clear”)
• A description that informs the AI how to behave
• Tags or content types that define when to use it

These can be layered. For example, combine:

• *“SEO Focus – UK Market”
• *“Tone – Playful but Trustworthy”
• *“Structure – How-To Article”

… and you’ll get something regionally optimised, on-brand, and campaign-ready - all without crafting a prompt from scratch.

This is less like typing to an assistant, and more like briefing a trained team member.

Why this matters

In real-world content ops, consistency is hard. With different writers, overlapping campaigns, and constantly evolving tone guidelines, quality often suffers.

While AI can scale production, it lacks guardrails unless you build them in - and that’s where Instructions shine.

“Instructions give you scalable constraints.”

You're not just generating content faster - you're generating it with intent. And that's the tangible application of agentic theory: shaping a system that understands your brand, adapts to context, and produces consistent, quality output.

Real-world examples from 26 DX

At 26 DX, we’ve been working with Instructions in live client work. Here’s how they’re already helping:

1. Brand consistency at scale

Reusable Instructions like:

• “Tone – Friendly and Plain English”
• “Voice – Authoritative for Public Sector Audiences”
• “Avoid – Hyperbole, jargon, or idioms”

...have drastically reduced review cycles and helped non-writers produce better first drafts.

2. SEO and content structure baked in

We’re using Instructions like:

• *“Use target keyword three times naturally”
• *“H1, H2 and bullet points only – skip long intros”
• *“Include a CTA at the end”

These help us keep structural consistency without relying on third-party tools or plugins - especially valuable for high-volume work like FAQs, blog summaries, and meta descriptions.

3. Multi-market localisation

Instructions also help with regional variants - from British/American spelling to tone or cultural preferences. It’s not perfect yet (nuance still needs a human), but it’s a solid head start for international content.

Prebuilt Instruction Agents

With Opal now standing alone, it's worth highlighting how prebuilt instruction agents are built directly on top of Instructions. These agents are ready-made behaviours designed for specific content use cases - and they’re available out of the box.

Some examples include:

• Campaign Brief – Generates structured campaign overviews
• Industry Marketer – Tailors a page based on an industry-specific need
• Keyword Research – Suggests keywords based on a seed term
• Marketing Researcher – Builds out campaign strategy
• Tone of Voice Sample – Helps define brand tone
• Video Transcription – Converts videos into text, subtitles, or translated output
• Experiment Plan and Personalisation Advisor – Focused on Optimizely Experimentation and personalisation strategy

See full list

Optimizely shares the underlying Instructions for each of these agents, giving you a clear template to work from when building your own. It’s a solid starting point if you’re looking to create something custom but don’t want to begin from a blank sheet.

Building, Sharing and Managing Custom Instruction Agents

As Opal matures into a standalone product, the real power comes not just from using the built-in agents, but from creating and managing your own custom instruction agents - reusable, governed sets of behaviour tailored to your organisation’s needs.

Once published, custom agents sit alongside the prebuilt ones in Opal, making them easy to discover and reuse.

Managing change at scale

As your content model evolves, so will your Instructions. To keep things consistent, Opal supports several approaches to managing and maintaining your instruction agents:

• Edit directly in the Opal admin panel, if you have access
• **Submit changes via your implementation or governance team
• Export and version-control Instructions as part of a centralised prompt library, especially useful for regulated environments or teams managing multiple brands

For teams managing a growing library of instructions and agents, Optimizely also offers tooling to help with:

• Import/export workflows
• Schema validation
• Environment-specific change control

Treat your instruction agents like any other content model - define them clearly, review them regularly, and version them where possible. That way, you get the scale and flexibility of AI, without sacrificing brand or quality.

Tools for Instruction Agents

If you're running a more advanced setup, Optimizely also offers tooling to support instruction agent development and lifecycle management. This includes:

• Import/export support \
• Schema validation \
• Change tracking across environments \

Explore the tools

These tools are especially useful for managing Instruction libraries across multiple brands, regions, or environments with change control in place.

Wrapping up

Opal’s Instructions might not shout for attention, but they’re quietly powerful. They shift your mindset from prompt engineering to behaviour engineering - defining not just what you want the AI to do, but how you want it to act.

As Opal continues to evolve as a standalone product, expect Instructions and instruction agents to play a growing role in how brands structure, scale, and systemise their approach to AI-powered content.

And for those of us aiming for more agentic workflows - this is where it gets real.