Some agents are live. Some are in flight. And some… we’re still wrapping our heads around.
As Anjin has evolved, so has the way we think about agents.
At first, it was about building useful tools - specific agents that could automate or enhance particular workflows. Then it became about modularity, chaining, permissions, and infrastructure.
Now, we’re starting to look forward.
What kinds of agents should exist?
What kinds of behaviours feel missing - not just in Anjin, but in the wider agent landscape?
Here are a few we haven’t built yet… but keep coming back to.
1. The Unblocker
What it does: Detects when a user is stuck, frustrated, or looping - then suggests context-aware next steps.
Why it’s tricky: It requires reading signals like hesitation, restarts, unclear goals… all without being intrusive or annoying.
Why we care: Great AI shouldn’t just answer questions - it should know when the real question hasn’t been asked yet.
2. The Observer
What it does: Watches agent chains passively, then logs insights and recommends optimisations.
Why it’s tricky: It needs to understand why a particular chain worked or failed - and how that might change over time.
Why we care: Most users don’t want to analyse logs. But they do want to know what’s working.
3. The Memory Architect
What it does: Helps users define what information should persist across sessions - and how.
Why it’s tricky: Persistent memory gets messy fast, especially with multiple agents. Context overload is a real thing.
Why we care: Long-term agent utility demands continuity - but it has to be intentional.
4. The Co-Creator
What it does: Builds new agents with you, based on conversation and use case discovery - not dropdowns and config panels.
Why it’s tricky: It’s not just a “template builder” - it’s an interactive partner that has to understand real needs and abstract them into executable logic.
Why we care: Most users know what they want help with. They just don’t know how to turn that into an agent. That’s a gap worth solving.
5. The Clean-Up Crew
What it does: Identifies agents you’ve stopped using, agents that are underperforming, or chains that are likely redundant.
Why it’s tricky: Declaring something “not useful” takes confidence - and nuance.
Why we care: As users build more, the system needs to help them refine, not just expand.
Why We Haven’t Built These Yet
Some are technically hard.
Some require UX patterns we haven’t solved.
And some are just ahead of where users are right now.
But they all point to something we believe in deeply:
AI agents shouldn’t just be functional. They should be supportive, reflective, and collaborative.
That’s where this is heading. Not “smarter” agents - but more thoughtful ones.
Want to Build With Us?
We’re always thinking about what’s next. But we’d rather think about it with people who are building, breaking, testing, and dreaming with us.
If there’s an agent you wish existed - or an idea you’d want to test - Join the community and pitch it.
Or, take a spin through the rest of the series to see how far we’ve come:
- Inside Anjin #09: Designing for Permission, Not Just Access
- Inside Anjin #08: Measuring What Matters
- Inside Anjin #07: What Happens When Agents Talk to Each Other
