Enterprise AI patterns

Leading & designing the way employees use AI on Meta internal tools

Meta's AI assistant was built for engineers, in a different org. Mine had none of its own. I founded the Enterprise Products AI Design Guild that gave it one, and built a chat pattern the company's design system adopted wholesale.

Problem:

No pattern language existed. Teams were reinventing the same AI interactions from scratch, and leadership support for fixing that stayed uncertain.

Outcome:

A founded guild, published design principles, and a chat pattern adopted as the company's foundation — still Meta's design-system guidance as of 2026.

Problem:

No pattern language existed. Teams were reinventing the same AI interactions from scratch, and leadership support for fixing that stayed uncertain.

Outcome:

A founded guild, published design principles, and a chat pattern adopted as the company's foundation — still Meta's design-system guidance as of 2026.

Role:

Founder & design lead

Timeline:

2024-ongoing

Role:

Founder & design lead

Timeline:

2024-ongoing

Flown in for a sprint to answer a question nobody owned

Search's Metamate partnership raised a question for my org: if Metamate wouldn't serve us directly, who was defining GenAI here?

I was flown to headquarters for an initial sprint. It produced a sticker sheet of patterns — the first design language Enterprise Products had — and won leadership sign-off right away.

The format failed before it found what worked

My first guild format was a forum — group readouts, open critique. Designers new to GenAI didn't want to expose half-formed work to a room, so I rebuilt it around one-on-one partnership and lightweight readouts instead.

Visibility, then priority

I embedded in the ML team's own meetings, seeing product ideas before teams had a designer assigned. I also audited which patterns products were actually using, so EPS could prioritize by evidence, not guesswork — chat came out on top.

One patternthe company adopted wholesale

Chat was the pattern every team wanted first.

Teams either built their own bespoke pattern, or stalled — reluctant to design an interaction language from scratch.

One fully specified pattern any team could pick up directly. Meta's design system adopted it as the company standard.

What mattered wasn't that a pattern existed — it removed the reason not to build at all.

Chat was the pattern every team wanted first.

Teams either built their own bespoke pattern, or stalled — reluctant to design an interaction language from scratch.

One fully specified pattern any team could pick up directly. Meta's design system adopted it as the company standard.

What mattered wasn't that a pattern existed — it removed the reason not to build at all.

The patterns a small library, not one component

Twelve patterns came out of this work, adopted across seventeen product-team experiments. Chat came first — these are a few others.

Results

12

Shipped patterns

17

product team adoptions in 2024

Chat pattern became basis for all internal chat experiences

Later extended into agentic patterns by our design systems team

Leadership recognition

My pattern work featured in a presentation that reached Zuckerberg's small circle

None of this looked obvious at the time. It reads as default now, the way good patterns eventually do.

Leadership support stayed ambiguous, and I rolled off as priorities shifted. Even so, this is still EPS's design system guidance as of 2026.

None of this looked obvious at the time. It reads as default now, the way good patterns eventually do.

Leadership support stayed ambiguous, and I rolled off as priorities shifted. Even so, this is still EPS's design system guidance as of 2026.

Jonathan LoPorto