Beyond the Chatbox

For the last two years, most “AI for work” products have shared the same shape: a chatbox for the prompt.

It’s understandable. The chatbox is a low-friction interface for a powerful new capability. You type, the model responds, and you get an immediate hit of usefulness. For a lot of tasks, drafting, summarizing, rewriting, vibe coding, it works.

But if you zoom out and look at how work actually happens, the chatbox is also a trap. Not because conversation is bad, but because conversation quietly pushes the hardest part of work back onto the user.

The hardest part isn’t generating words. It’s running the loop.

It’s knowing what matters today. It’s deciding what should happen next. It’s keeping momentum alive across stakeholders. It’s remembering the detail someone mentioned three meetings ago that suddenly becomes relevant. It’s noticing the deal is cooling down before it’s obvious. It’s catching the mismatch between what was promised and what’s actually happening.

The chatbox doesn’t solve those problems. It just gives you a powerful new way to do more internal pieces of work.

The hidden tax of the chatbox

A chat interface assumes something that usually isn’t true in real life: that the user knows what to ask for and has the ability to describe it precisely in a way the AI can best understand.

To get value out of a chatbox, you need to do at least three things every time:

1. decide what the real problem is,
2. translate that into a prompt, and
3. provide enough context data for the model to respond correctly.

That sounds manageable, until you realize this is exactly the same “glue work” people are already drowning in. You’re still carrying the situation in your head. You’re still the one stitching together context. You’re still the one deciding what’s important. Now you’re just doing it with a smarter tool.

This is why so many professionals have the same experience: AI makes them faster at producing artifacts, but it doesn’t make the day feel lighter. It doesn’t reduce responsibility. It doesn’t remove fragmentation. It doesn’t stop work from slipping through cracks.

It’s not that the model isn’t intelligent enough. It’s that the product is asking the user to be the real operating system.

Why “better prompting” is the wrong direction

A common response is to treat prompting as a new skill. People share templates, playbooks, agent chains, prompt libraries. Some teams even formalize “prompt ops.” And yes, this can help in the same way that learning keyboard shortcuts helps.

But it’s not a fundamental solution. Because real work is not a static input-output problem.

Work is dynamic. Priorities shift mid-day. Stakeholders change their minds. A message you get at 4pm can invalidate what you believed at noon. A project that seemed stable becomes urgent because a dependency moved. A deal that looked healthy becomes risky because the champion went silent.

In that world, asking users to write prompts and maintain workflows is like asking them to hand-code their own operating system every morning. Even when it works, it doesn’t scale. And it certainly doesn’t feel like relief.

The real shift: from “ask and answer” to role-based managed work

The next interface won’t be “chat, but smarter.”

It will be a system that continuously understands your situation and actively drives it forward. Instead of waiting for you to ask the right question, it will surface what changed, what matters, propose next actions, and help close loops.

In other words: the AI should keep writing the prompts on the user’s behalf, and partner with them to drive loops to closure.

If you look at the moments where work actually breaks down, they almost never look like “I didn’t have a good enough draft.” They look like:

• “We didn’t follow up in time.”
• “We forgot that someone had a concern.”
• “We assumed ownership, but no one actually owned it.”
• “We were aligned in the meeting, but not aligned afterwards.”
• “We missed the real decision-maker.”
• “We didn’t realize the plan had changed.”

These are not writing problems. They are context and momentum problems.

A managed-work system starts from a different premise: your job is not to constantly query a model. Your job is to make decisions. The system should do the context-holding and the loop-running.

So instead of, “Summarize this thread,” you get something closer to: “This conversation created a commitment with no owner. Here’s the recommended owner, here’s the message to send, here's action to take, and here’s the follow-up checkpoint.”

Instead of, “Help me write a follow-up,” you get: “This stakeholder went quiet after you mentioned pricing. Similar patterns usually indicate an internal approval loop. Here are the three most likely blockers and the single question that will surface the real one.”

Instead of, “What should I do next?” you get: “Two deadlines shifted, one dependency is at risk, and there’s a narrow window to salvage momentum with this partner. Here’s the sequence of actions to take today, and I can execute steps one through three.”

That’s not magic. That’s what a AI system looks like when it is operating with continuity with your context graph rather than working in isolated chat sessions.

Why proactive beats reactive

There’s a reason high-performing leaders don’t run their day by waiting for interruptions and responding one by one. They run their day by scanning for what matters and keeping the highest, leverage loops moving.

Most work tools are reactive. They deliver messages. They don’t deliver momentum.

A proactive system flips the shape of the experience. Your default view isn’t an inbox of noise. It’s a feed of “what’s actually in motion” and “what actually needs attention,” grounded in the way your work evolves across time.

This matters because attention is the scarcest resource in modern work. A system that requires you to constantly remember what matters is not a productivity system, it’s a tax collector. A system that reliably surfaces what matters is leverage.

The end state: less orchestration, more outcomes

The goal isn’t to eliminate conversation. Conversation is part of work. The goal is to stop forcing people to be full-time orchestrators of their own context. Because that’s where the cost is hiding.

When your tools don’t hold context, you do. When your tools don’t run loops, you do. When your tools don’t maintain momentum, you do. And that’s why even very capable teams can feel like they’re working hard just to stay in place.

A chatbox makes you a more powerful individual contributor inside a fragmented system.

A managed-work system changes the system.

It reduces the need for you to write prompts, define and maintain workflows, or constantly re-explain your situation. It turns “keeping things moving” from a human memory task into a system behavior.

That’s the shift that actually matters. Not a smarter answer, but a different responsibility model.

The AI-native era won’t be defined by who built the best chat interface. It will be defined by who built the systems that can carry work forward, quietly, continuously, and reliably, so humans can do what humans are best at: judgment, taste, strategy, and relationships.

That’s what we should demand from our tools now. And it’s the direction we’re building toward.

— Ethan Founder, Meland Labs