Show HN: What should the GUI for AI agents look like?

I'm glad folks are experimenting with different paradigms.

That said, Hacker News might not be the best audience for this informal survey. I'd wager at least 70% of the people reading this are still interacting with their computers using "strange terminal commands", by choice. We aren't representative of what normal people will use or prefer.

I completely agree with you that the chat interface is very undercooked and there's a big untapped space for agent-first interfaces, perhaps even replacing traditional desktop metaphors. You're spot on with your premise.

With that said, from the material you show in your homepage, it still seems like you're doing something very similar to OpenAI and Anthropic's first party apps? I think it would help your pitch if you could show some task or workflow where your interface truly makes a difference; loading files and asking questions about them is a staple of most harnesses.

Best interface in my opinion is a git tracked folder, files as state, agents coming in and doing work. It's not just a chat interface because in chat mode everything flows like water under the bridge. It is files as agents, where you get the agent to write a task in a file, and then it updates the file like a blackboard as it performs work. Same task file can be seen by worker agent, multiple judge agents (for plan and implementation), documentation agents and reflexion agents (improve workflow). I would also track every user message in a chat_log.md for reflexion agents (what is wrong with the workflow) and intent review (is the agent still following user intent?).

Creative take, I like the visual representation on infinite canvas a lot. Feels like freedom. What doesn't feel like freedom nor future is the manual selection of tools. AI should naturally pick the right tools out of all possible tools, choosing what abilities or data it can have each time is hassle. What is the reasoning for manual and explicit tools? Still, very cool UI :-)

Drop the "G". The best UI for AI Agents is a physical UI (not a graphical UI). Do something like PikaCube [0] and disrupt the mouse.

[0] https://pikacube.pages.dev/

If you are taking inspiration from OSs, consider creating a new AI-based OS rather than a tool.

I want to see a graph or flame graph or something akin to how Temporal renders workflows in real-time. I want to follow along as all the different sub-agents and tool calls do their work. I also want to know how much this is costing me at each step.

Just like with a Temporal workflow, I don't care how things work 99% of the time. But that last 1% when I really need to dig in and understand how and where things went sideways this type of view would be invaluable.

I believe, fundamentally, that interacting with AI will increasingly move in the same direction as interacting with coworkers.

Slack, video/audio meetings, shared screens, paired programming, cooperation in the classical sense. I expect humanoid robots to be a standard part of life by 2050, so then they'll be able to complain about the coffee at the office like everyone else, I guess.

I think there is going to be an intersection between a personal operational dashboard like this and the UI for an AI Company.

I've been predicting for awhile that 2026 is the year of the AI Employee (or serious attempts) and 2027 is AI Companies.

I have been seeing RFPs recently asking for end-to-end solutions to operate companies in particular industries with the goal of day-to-day only requiring infrequent human-in-the-loop.

Anyway, if you think about what for example some professionals might use a Marble like tool for, say managing their freelance business. You might want to create a task and then promote it to be scheduled and then have that artifact like a report have a dedicated place on the screen.

The difference with a vertical is that you will know a lot of the reports and data types you need ahead of time. The advantage with just prompting things dynamically as you go is you get exactly what you need and don't have to sort through a bunch of features that aren't important to you.

Even though a business vertical is qualitatively different, you still might use the same system to build it. Maybe something like a self organizing AirTable. A set of projects as departments.

Sorry I started rambling nevermind.

Lost me at having to click to use particular tools

When I initially saw this I was excited because what I’m trying to solve for is visibility of both semi and fully autonomous and agents at runtime. Marble is an improvement over chat and Claude but it’s incremental IMO.

We’ve toyed with a meta agent style interface which you chat to do everything. Add tools, connect datasets, build agents, deploy and schedule them, build skills, etc.

But at some point agents have to have an interface for human approvals, validation, training, and tuning.

I guess it’s like a spreadsheet. Ultimately agents are part of custom apps which are each going to be as unique like snowflakes. So it’s difficult to have one standard abstracted runtime interface without feeling obtuse to a business user. I think what I’m describing and wanting are new app layer building blocks vs one interface to rule them all.

All of this said, I’m all for a better personal llm harness interface (which is what this is). So keep up the great work guys!

I think it should look like this: https://www.dreamcoder.ai/

A smart, tiling window manager with workspaces and a great tool harness. An everything app, if you may; bring the terminals, code reviews, etc. into the tool itself, and make it high fidelity: no blocky terminals for the app itself.

I think my outlook client might be the most ideal interface once we get some of the rough edges sorted out.

For my autonomous agents, I'm using Github Issues as the UI. And yes, I know this is a total abuse of what a github issue is supposed to be. But it's better than spinning up an ephemeral web site, or hooking into Slack or Telegram (at least for now). We'll see how long it works or if I think of something better.

> Before GUIs, interacting with a computer was limited to strange terminal commands [...] Even though the strict syntax has been replaced with natural language

This post, and particularly the two quoted sentences, has activated anger pathways in my brain that I didn't know existed... An extremely visceral, primary, response. I find it unbelievably jarring to read.

this looks good! I bet with a bit of polishing and iteration, it is gonna be pretty good. A use case that comes to mind is learning. I create different custom setups for learning all the time. It feels like this could be a nice way to do it.

theoretically this is interesting, but in practice it's just one layer that sits atop what is ultimately still many individual chat boxes. Did you try anything else, anything a little more novel?

These Todo items on the side in the video seem to be useful.

I'm not so sure about that tool toolbar. It might as well just be a tool selector dropdown in a typical chat window.

A quick FYI: at marbles.com/learn, starting from „Turn Messy Notes into Today’s Task List“ the demo videos don’t work for me. I tried Safari and Chrome.

This is definitely an iterative improvement on what we have today. However, I think from a fundamental approach, you've just highlighted the pain of working with AI that I hadn't quite been able to verbalise until seeing this.

Go and have a look at the history of Microsoft OLE and OpenDoc and you'll see some of these ideas have been around for a while. When these ideas were first being shared in the industry (and yes, I am that old), here's the story that was being told:

In the future, you wouldn't buy "a word processor" or "an image editor", you'd buy components that could do specific tasks, and you'd bring them to your document/work as and when you needed. I'd have DTP components for layout, word processing components for spell checking, image components for resizing and colour balancing, and I'd be heads down in my fancy newsletter or report, and I'd be bringing functionality to the work I needed to do, without having to context switch and import/export things by hand.

The demo you've shown on the homepage is almost the opposite of this: I'm not bringing things into my work (fanning in), I'm starting agents to go and do lots of different bits of work (fanning out), that eventually I'm going to need to think about how to structure and fan in again.

I think coding interfaces kind of force us into a fan-in: here's my code base, now I want to improve test coverage, now I want to refactor, now I want to add this SDK, now I want to change this logic to use the SDK based on configuration params, now I want to improve my integration tests, and so on. I'm flowing through my work in a single asset (the code base), and able to bring agents and other tools to it.

Now, I'd like that, but for everything else. The UI I'd like allows me to build around the assets I am building.

Planning a trip? OK, I need to bring various agents along to help me plan an itinerary (in my calendar), based on location, budget, and preferences. It's not good enough to list the top 10 must-see places in Venice, I need them to fit into my calendar including travel time, and my budget - I want agents to help me plan that out. Ideally I want agents to help me and my partner both contribute to that based on our wants.

Need to do a report for the big boss? OK, I need to start with an empty report doc, and bring various sources of data, and take it through transformations to build the narrative/deck I need to write. Here's some excel data, here's an agent that can do some competitive analysis, here's another that is pulling quotes from real customers on social media, and so on...

Want to learn a new skill? We're going to need to have a framework for assessing where I am in terms of that skill (the "single asset" we're working on is "what I know"), and then agents to figure out the best way for me to fill gaps and reinforce what I already know.

Your demo is nicer than a chat box, but I think you need to think about the job that needs to be done, and how to bring agents to that job (perhaps even suggest those jobs - that could be another agent), not build a UI to just allow me to fire off dozens of agents that then have output sat in multiple windows across my screen.

Very interesting idea, I actually wonder if you are not going far enough?

See this video from the lovely YouTube channel Interface Studies: https://www.youtube.com/watch?v=ARjisP9enZ4

I'm generally very aligned to the thinking in that video. AI offers an opportunity to take a massive step-back and re-evaluate what interfaces are in general.

IMHO Douglas Englebart had a very compelling and beautiful vision for what computers could do for us, make us better people and better thinkers.

If MarbleOS helps to re-surface that conversation, to my mind that's a good thing.

The demo video is a great example of one thing I have been saying for ages - the quality of the LLM output depends heavily on the communication skills of the user.

Some people say AI is great and some people say AI is terrible - but how can anyone compare these things because they depend so much on the skill of the user.

Case in point - the demo video. The user says "find me all the HCI articles on hacker news" and then a research task started... ok great so far.

Then the user says "they should be made into slides"

What exactly do you mean by this? Do you mean "from the set of results you found, narrow them to be only the ones that have been made into slides" or does the user mean "I would like you to then take those results and make them into slides"

There will be people who read this and say its obviously the former and people who read this and say its obviously the latter, and neither of those is the point I am trying to make, the point is that its very ambiguous and if the AI did the "wrong" one, then the user would complain that AI is garbage but the real problem is that the instruction was very unclear because the users communication skill was poor.

It does feel novel.

However, I believe the GUI for AI agents that will win is a plain text editor with a file tree, a terminal pane for coding (or preview pane for other kinds of tasks) and a chat sidebar.

Even if the code editor is not used to type, it's there for psychological safety. It lets you inspect and navigate what is being created. Has tremendous value.

Plain text has survived countless software revolutions, so it's likely to become the dominant format for artifacts produced by AI. Non-plain text things are likely to adapt instead of the other way around.

"Strange terminal commands" have survived, while even poweruser toolbar style have started to fade (I love them, but they're fading away). It clearly has some value (hint: speed over familiarity).

I think the poweruser toolbar style was indeed the only thing in the past couple of decades that really tried to match the terminal for speed of interaction. In my opinion, it succeeded, but it has its limits too (tends to clump with stuff over time; has a limit on the amount of abstractions it can hold). I would love to see a candidate GUI that tackles that speed angle.

I don't think we should homage GUI for AI agent workflows. The terminal and Mac GUI are 100% deterministic when you click an icon, but I'm not sure if visualizing agent workflows is the right call.

The problem is that AI workflows are inherently different for each person.

The current approach feels like it's forcing a CLI-based model on users. I also don't think chat is a suitable fundamental unit for task delegation.

I've worked on writing a compiler using both hand-written code and AI. Once the project exceeds a certain size, the chat itself becomes a bottleneck. In those cases, I needed proof and gates like Rocq. So in essence, AI coding requires clear negative gates that should be rejected when approaching the goal.

I think something like Figma's canvas model might become the new interface.

Because no matter how meta you make it, agent workflows are ultimately optimized for the individual. My settings often don't match someone else's.

Computers also have this problem, but in that case, you can enforce defaults. With AI agents, it's different.

That's why I think we need a canvas—something like a workspace next to the user space where agent workflows can be dynamically adjusted.

I focus most of my coding effort on building gates that AI-generated code must pass through when it's being produced. That approach has allowed me to handle much larger volumes of code.

The future agent GUI will be about supervising multiple asynchronous tasks.

In a way, this might end up resembling object-oriented programming—just with task graphs instead of object graphs.

Once AI starts generating code, it flows out like water through a burst dam. It's impossible for any human to fully understand it all. At first, I tried to understand every line, but that actually turned out to be less efficient than just writing the code myself. There's a clear fundamental mismatch between the way AI thinks about code and the way I do.

This mismatch seems like an unsolvable impedance mismatch problem—similar to the one between ORM and SQL. So once you decide to use AI agent code, you have no choice but to shift your focus from reviewing the code itself to trusting the gates you've built around it.

The key question is how tightly you can build those gates. And I don't think chat-based interfaces are capable of providing that level of control.

I only use web chat with copy&paste and file upload option. Anything else I don't trust, especially since the people who create AI harnesses and apps use AI to develop them.

Xerox Parc is one of my favorite anecdotes. However it's not for UI, it's for how major companies like Xerox kill innovation by trying to make it fit to the form factor of their business.

A mouse and screen with a folder system was a way of humanizing the command line. The funny thing is as a developer, if I know what I want to do on the CLI it's faster than using a mouse and keyboard, because it's essentially shortening the length of time it takes me to communicate a complex idea to a computer and have it execute that.

The chat interface sucks, but it is pretty close to the ideal way of interacting with a computer. Talk to it. It shows you something. Talk to it again. That paradigm is pretty close to how we interact with humans. Talking and showing.

The reason why the UI sucks for AI right now is because AI sucks.

You shouldn't have to manage it as much as you do. But we have to manage it delivers about 90% of what it promises with the missing 10% being critically important.

It's extremely hard to create human bonds with a dementia patient. It's hard to hire an engineer that can't do math.

Solving memory for AI as well as non-stochastic output will leapfrog all of the problems most AI engineers are dealing with right now.

I generally build stuff with AI for 3 reasons: It solves a real problem right now, it's adding to some fundamental problem around the discipline, it's fun.

AI harnesses are pretty big right now, but they will likely get leapfrogged by a better algorithm. At the end of the day the best "harness" for AI is just trust.

Do we really want a complex harness for AI to manage it in a more time consuming way, or do we want to just trust that it can follow instructions and manage complexity on it's own.

The competitive issue around this is that everyone is working on AI right now, so it's realistic to expect a business landscape where harnesses and fundamental shifts in AI capability take turns leapfrogging each other. Temporary solve to current limitations. Paradigm shift in model behavior. It's behaving similarly to moore's law, but on a much faster scale that is harder to measure discretely.

In the spirit of Xerox Parc, I think the bigger innovation isn't reinventing a few things that we already have such as an OS and a chat interface and file management. But rather looking at things the same way those engineers did, which is what does the future of human interactions with machines look like.

If my computer actually had a brain, I would just talk to it, and it would show me things. It might 3D print something so I could try it in the real world. It might show me some things I could have shipped to my house. It might design a UI for me, pull some data etc. I wouldn't have to manage it because it would have access to a lot of fundamental business problems and solutions.

The issue with that way of interacting with computers is that's centralizing knowledge inside of a computer. Doing that erodes competitive business advantage. The two edges of the political spectrum where that works is monopolistic business practices and communism.

And that's actually what you're seeing a gigantic war over right now in AI. China pushing open weight models for free to undercut US AI dominance which requires an insane capital expenditure to keep afloat.

Realistically, I don't think the answer to a lot of these issues is to build a business. I think a lot of businesses built around AI will become irrelevant in months. Open claw is something that 10 years ago would have become a unicorn company. But it got eclipsed by a flood of agent harnesses. And Open claw itself was a leap frog of cursor which was less than a year old. Cursor itself now a part of SpaceX's trillion dollar play.

I just don't see software as a service being anywhere close to as lucrative as it used to be.

It's way more realistic to assume that China is going to undercut US AI prices until they find a way to run complex models on small amounts of compute, eventually wiping a significant amount of value from the US market. At the same time, a lot of the models like OpenAI and Deepseek will be using user input as training data to develop future projects.

It's hard to say when that will happen, but there's a realistic future where AI just cobbles together scalable performant SaaS in rust while running on a micro PC.

The bigger use problem at that point is that most people are really bad at making products. They're really bad at identifying problems, solutions, and use cases for things. Most people end up building one off solutions for temporary problems. Which is fair, because most people aren't engineers or product designers.

I suppose the biggest question in AI dev right now is something like: how can you in any way, quickly tell what you just did with an AI agent in a way that's visual and easy to understand. People call AI stuff slop because it spits out a bunch of output and people can't easily tell what the output is or if it does what it's supposed to. That's a pretty big problem.

It looks like you're building for folks who don't want to dive super deep into AI? In that case I'd gear your product more towards a "thing doer" rather than a complex thing you can crack open and get under the hood.

You generally solve that problem by doing the Amazon method. Start by selling books. Move to selling everything.

So the question would be, what's your Xerox Parc level belief on humans and computers and AI and would would be the first problem you'd solve using that paradigm.

Personally I feel like it's way more efficient to start building social groups right now instead of companies and products, because a lot of what we did on the scale of web 2.0 is just going to be trivial.

[flagged]

[flagged]

[flagged]

[flagged]

[flagged]

[dead]

[flagged]

Why are you assuming that the human is in charge? Needing a human to drive the AI is probably a transitional phase.