AI Brand assistant

Every designer wants autonomy. Every company wants consistency. And every slide deck seems determined to sabotage both.

At Workable, our brand design team was constantly asked to “just brand this real quick,” which really meant: rebuild layouts, reapply color codes, rewrite intros, and do it yesterday.

We didn’t need another Figma library. We needed a smarter teammate.

So we built a custom GPT-powered brand assistant. Not a glorified wiki or chatbot, but an intelligent system that could actually apply our brand logic and help teams stay consistent without constant hand-holding.

We started scrappy. Using ChatGPT builder, Make.com, and some duct tape, we connected a webhook to a GPT instance. The assistant could ask questions, generate presentation copy, select images, and, after a round of user approvals, trigger a Google Slides automation to send out a branded deck.

It worked. But it was fragile. Debugging was a mess, the UX was clunky, and switching platforms disrupted the flow.

We migrated everything to n8n, built a custom MCP server, and transformed the assistant into a modular AI agent. This meant conditional logic, real-time adaptability, and no more context-switching for users.

Now, from a single prompt, the assistant could:

Interpret the request
Generate a fully branded Google Slides deck
Return a live link directly in the chat

Bonus: We could log requests, debug intelligently, and even let the agent debug itself.

We then built a custom interface in Framer and overlaid it onto our new brand site. Unlike fake demos, this was the real thing, fully integrated with GPT, styled with our brand, and complete with type animations.

Once stable, we leveled up:

Integrated Figma’s MCP to export branded components on demand
Extended support to onboarding decks, sales kits, and social templates
Began testing text injection into exported assets (WIP but promising)

This wasn’t about hype. It was about agentic AI systems that don’t just respond but understand the logic behind your brand.

Anyone could say “Here’s a product update” and get back a well-structured, well-branded deck that sounds and looks like Workable, not because the assistant is guessing, but because it was trained to think like our team.

Results and Impact

No big launch. No dramatic unveil.

Just quiet adoption, curious questions, and a lot less “Can you brand this?” messages in my inbox.

Some bugs? Sure. But what we gained was a scalable way to preserve brand integrity, without bottlenecks.

This assistant became the connective tissue between static guidelines and living systems. It didn’t replace designers.

It let them focus on the good stuff.

NEXT CASE STUDY

Workable Website Redesign

Workable’s marketing website had grown fast, but not always strategically. We needed more than a visual facelift. We needed a rearchitecture of how we communicated who we are, what we offer, and why it matters.

→

AI Brand assistant

We didn’t need another Figma library. We needed a smarter teammate.

So we built a custom GPT-powered brand assistant. Not a glorified wiki or chatbot, but an intelligent system that could actually apply our brand logic and help teams stay consistent without constant hand-holding.

We started scrappy. Using ChatGPT builder, Make.com, and some duct tape, we connected a webhook to a GPT instance. The assistant could ask questions, generate presentation copy, select images, and, after a round of user approvals, trigger a Google Slides automation to send out a branded deck.

It worked. But it was fragile. Debugging was a mess, the UX was clunky, and switching platforms disrupted the flow.

We migrated everything to n8n, built a custom MCP server, and transformed the assistant into a modular AI agent. This meant conditional logic, real-time adaptability, and no more context-switching for users.

Now, from a single prompt, the assistant could:

Interpret the request

Generate a fully branded Google Slides deck

Return a live link directly in the chat

Bonus: We could log requests, debug intelligently, and even let the agent debug itself.

We then built a custom interface in Framer and overlaid it onto our new brand site. Unlike fake demos, this was the real thing, fully integrated with GPT, styled with our brand, and complete with type animations.

Once stable, we leveled up:

Integrated Figma’s MCP to export branded components on demand

Extended support to onboarding decks, sales kits, and social templates

Began testing text injection into exported assets (WIP but promising)

This wasn’t about hype. It was about agentic AI systems that don’t just respond but understand the logic behind your brand.

Anyone could say “Here’s a product update” and get back a well-structured, well-branded deck that sounds and looks like Workable, not because the assistant is guessing, but because it was trained to think like our team.

Results and Impact

No big launch. No dramatic unveil.

Just quiet adoption, curious questions, and a lot less “Can you brand this?” messages in my inbox.

Some bugs? Sure. But what we gained was a scalable way to preserve brand integrity, without bottlenecks.

This assistant became the connective tissue between static guidelines and living systems. It didn’t replace designers.

It let them focus on the good stuff.

Workable Website Redesign

Workable’s marketing website had grown fast, but not always strategically. We needed more than a visual facelift. We needed a rearchitecture of how we communicated who we are, what we offer, and why it matters.