The Missing Layer in the Agent Stack: Design

By Mu Li · March 21, 2026

Software is changing shape.

For a long time, most tools were built for one kind of user: a human sitting in front of a screen, clicking through a workflow step by step.

That assumption is starting to break.

Today, more of the stack is becoming agent-driven. Coding is becoming agentic. Research is becoming agentic. Execution is becoming agentic too.

You can see it in the way tools are evolving:

• Claude Code is not just autocomplete in a nicer shell. It is an agentic coding tool.
• Cursor now talks openly about agents that turn ideas into code.
• Codex is moving toward multi-agent workflows and parallel software work.
• Google Stitch is pushing UI generation into a more AI-native design workflow.

We are no longer only using tools ourselves.

We are starting to work with AI.
And increasingly, AI is starting to use tools on our behalf.

That changes the interface.

The new “user” of many tools is no longer only a human.
Sometimes, it is an agent.

And most tools still are not built for that.

AI can generate. But generation is not design.

AI can already generate almost anything.

A landing page.
A dashboard.
A mobile UI.
A React app.
A full website scaffold in one shot.

That part is no longer the surprise.

The difference between AI-generated output and intentional design: while AI can handle structure (left), design intelligence helps create a more composed, high-end experience (right).

What still feels unfinished is what happens after generation.

The output may be fast. It may even be functional. But very often it still lacks structure, hierarchy, rhythm, restraint, and visual coherence.

It works.
But it does not feel designed.

Lokuma acts as a reasoning layer—planning, refinement, typography, consistency, and design system thinking—so outputs feel intentional rather than noisy.

That gap shows up everywhere once you start looking for it.

A page has all the right sections, but the visual weight is off.
A hero is technically fine, but the typography has no tension or clarity.
A layout is clean enough, but everything feels equally loud.
A product interface renders, but it does not feel composed.

This is the part many AI products still flatten.

Generation solves abundance.
Design solves organization.

Generation gives you output.
Design gives that output intent.

Why this matters even more for websites

Websites make this problem obvious very quickly.

A website is not just code. It is not just content. It is not just “something that renders.”

A good website has to do several things at once:

• communicate clearly
• establish trust quickly
• organize information with hierarchy
• feel visually consistent
• guide someone toward action

That is why building a website is still a design problem, not just a generation problem.

A lot of AI website outputs today are impressive for about ten seconds.

They are fast.
They are usable.
They are often better than a blank canvas.

But many still feel generic, overfilled, or visually unconvincing. The page exists, but the experience does not quite land.

If AI is going to build more of the web, it cannot stop at “good enough to render.”

It has to get better at how things come together.

What Lokuma is building

At Lokuma, we think this new web needs to be built for both humans and AI.

On one side, that means an AI website builder — a product that helps people go from idea to site much faster, without giving up structure, clarity, and taste.

On the other side, it means something deeper: a design intelligence layer for AI agents.

Install the Lokuma Design Agent via CLI to bring design intelligence into agentic coding workflows in tools like Claude Code, Cursor, and OpenAI Codex.

That is the idea behind Lokuma Design Agent.

Instead of treating design as a last-mile polish step, we think of it as a reasoning layer.

A layer that helps agents make better decisions about layout, typography, hierarchy, spacing, and visual balance.

In simple terms:

• your AI can handle logic, generation, and task execution
• Lokuma helps shape how the result is structured and presented

Lokuma Agent Designer sits alongside coding agents as a design intelligence layer—turning prompts into structured, visually coherent results.

This is why we do not see Lokuma as two disconnected products.

The website builder is one expression of the system.
The design agent is another.

One helps humans build faster.
The other helps AI build better.

Both come from the same belief:

design should not disappear in an AI-native workflow. It should become part of the infrastructure.

Agent Designer by Lokuma bridges raw generation and polished UX, with integrations for OpenClaw, Claude Code, Cursor, and OpenAI Codex.

From website builder to design infrastructure

A lot of website builders were designed for a world where the human was always the direct operator.

Choose a template.
Drag some blocks.
Edit the copy.
Adjust spacing.
Publish.

That model still matters. It is not going away.

But a second model is emerging beside it.

In that model, the human increasingly works through AI. The agent handles more of the production work. It writes, structures, edits, tests, and iterates. The human becomes more like a director than a pixel-level operator.

When that happens, the role of design changes.

Design is no longer only something a human manually applies through a tool.

It becomes something the system itself needs to understand.

Not as decoration.
Not as a skin.
Not as a final pass.

But as part of the logic of making something feel complete.

That is the shift we care about.

Why now

Because we are entering an agent-first stack.

There are already stronger layers emerging for code, research, and execution. Agents can already do a surprising amount of real work across those domains.

But design is still underbuilt.

There are many tools that can generate UI.
Fewer that can reason about composition.
Even fewer that can help an agent make outputs feel intentional.

That is why this moment matters.

The more software gets built by agents, the more obvious the missing layer becomes.

Without better design judgment, AI output tends to collapse into the same pattern:

• too much information
• weak hierarchy
• repetitive layouts
• inconsistent spacing
• generic visual decisions
• interfaces that function, but do not feel finished

The web is about to contain a lot more AI-generated surfaces.

The question is whether those surfaces will feel disposable, or designed.

Why us

Because our team comes from the overlap.

We have spent years around design tools, design systems, product building, growth, and AI workflows. We care about the part that is easy to dismiss but hard to fake: why some interfaces feel clear, calm, and inevitable, while others feel noisy even when they are technically correct.

That difference is not magic.
And it is not just taste.

A lot of it comes down to structure, hierarchy, proportion, sequence, and how decisions relate to each other across a page.

We think those principles can be translated into something AI can actually use.

That is the work.

What is still missing before AI can really ship something complete

If your AI can already write code, call tools, search documentation, and execute multi-step tasks, what is still missing before it can ship something that feels genuinely complete?

From our point of view, at least a few things:

• stronger layout reasoning
• better hierarchy decisions
• more reliable typography
• more consistent spacing and composition
• a better sense of what to emphasize, simplify, or remove
• a deeper understanding of design intent, not just design patterns

This is where we think the next layer needs to be built.

Not another generator.
Not another template engine.
Something closer to design judgment.

Lokuma’s view of the next web

We do not think the future is “AI replaces websites.”

We think the future is that websites — and the products behind them — are increasingly built through agentic workflows.

Humans will still set direction.
Humans will still decide what matters.
Humans will still recognize what feels right.

But more of the making will happen through AI.

And when that happens, design becomes more important, not less.

That is why Lokuma is building across both surfaces of the problem:

• an AI website builder for humans
• a design intelligence layer for AI agents

Same brand.
Same direction.
Two interfaces for the same shift.

The web is not just becoming AI-generated.

It is becoming agent-built.

And when that happens, design stops being a layer of polish.

It becomes part of the stack.

---

Further reading

If you want to explore the tools shaping this shift, here are a few worth looking at:

• Claude Code
• Cursor
• OpenAI Codex
• Google Stitch

And if you are building with AI already, we think the real question is no longer whether AI can generate.

It is whether it can generate something that feels whole.

That is the gap we are working on at Lokuma.