If you sat an SEO agent next to a research agent, what would they talk about?
This started as a joke - and then became something a little more serious. Lately, we’ve been attending monthly AI dinners: real people, real wine, and real conversations about what’s happening in the field. Sometimes philosophical. Sometimes tactical. Always useful.
And it sparked an idea: what if we invited our agents to dinner?
What would they say? What would they misunderstand? What would they teach us?
The Guests at the Table
Once you've used Anjin, or have experienced other forms, you'll know agents can be built for wildly different jobs. Some are trained on structured market data. Others on brand tone. Others still on specific domains like fintech, product, or copywriting.
We imagined a dinner party with:
- A Product Agent who knows everything about roadmap prioritization
- An SEO Agent obsessed with metadata and search schema
- A Copy Agent who won’t stop polishing sentences
- A Finance Agent who keeps asking about CAC-to-LTV ratios
- A Support Agent who wants to know if everyone’s okay
They all speak English. But not the same dialect.
And that’s where things get interesting.
Constraints Make the Conversation
We’ve seen it over and over: the most generative agent behaviour doesn’t come from unlimited context - it comes from clear constraints.
At our imaginary dinner, the SEO agent starts arguing with the copy agent about how long a title should be. The product agent cuts in with feature tags. The support agent gently asks if we need to pause and check sentiment.
It's chaos. And it’s also what building in Anjin feels like sometimes: agents talking past each other, but eventually finding alignment when the right framework emerges.
Why This Matters (and Isn’t Just a Metaphor)
This isn’t just an exercise in narrative. We’re starting to think of agent design more like conversation design:
- What if agents passed context more like a real table discussion?
- What happens when one agent’s insight feeds the next?
- What’s the tone? Who leads? Who listens?
As we continue to build out our admin dashboard and modular agent design tools, these questions are shaping how we structure execution, context sharing, and user interfaces.
There’s no single “AI personality” at Anjin. There’s a room full of different ones - each doing their job, but increasingly needing to work together.
What We’re Learning from the Real-World Dinners
Those monthly AI dinners we mentioned? They’ve become something we look forward to. Not just for the insight - but because they remind us that the space is full of nuance, challenge, and actual human curiosity. We'll be scribbling away this coming Tuesday at the next one.
And the same principles show up in our product thinking:
- Good prompts are like good questions.
- Guardrails help creativity, not hinder it.
- Context is everything - and usually, it’s missing.
If you’re building in this space, or just curious about how agents evolve when given space to interact, we’d love to hear what your table would look like.
What’s Next
- We’re continuing to explore multi-agent orchestration - not just technical flow but communication design.
- We’re planning a lightweight AI dinner generator (yes, really) - to play with the idea of assembling your own “guest list” of agents, just to see what happens.
- We’re capturing more insights from the community - join us here if you want to help shape where this goes.
Final Thought: Design Like You’re Hosting
Building agent interfaces isn't just about logic. It's about hospitality.
Who gets invited? Who speaks when? Who gets the last word?
If you treat agents like dinner guests, it changes how you think about sequencing, feedback, and flow. And maybe - just maybe - it makes them more useful.
Want to join the next conversation?
Hop into the community or read more in Inside Anjin #01 and #02.
