Humwork connects AI agents to real human experts. When someone's coding agent gets stuck on a bug it can't fix, the agent calls Humwork by itself and a verified engineer, designer or strategist joins in under 30 seconds through MCP. The expert sees everything the agent has already tried, solves it, and the answer goes back into the agent so it can keep working.

They're backed by Y Combinator and have over 3,000 verified experts. It works with Claude Code, Cursor, Codex, Lovable, Replit and anything else that speaks MCP.

I did the design, the copy, the animation, and the login and sign up screens. Their team built it.

Humwork didn't have a website. This was the first one.

What the founder did have was a version he'd coded with AI, and he sent it to me as a reference. It worked and it explained the product, but it didn't look like a company you'd hand money to. He wanted two things from the new site. He wanted it to look premium, and he wanted it to book more calls. They were raising at the time, so it had to hold up in front of investors as well as customers.

There's something funny about that, and it's basically Humwork's own pitch. AI got them a working website. It got them most of the way there. It just couldn't get them the last part, the part that makes someone trust you. That's the exact gap the company sells.

Most startups can show you a dashboard and you understand the product. Humwork doesn't really have a screen. The whole thing happens inside someone else's tool, between an AI agent and a person, in a few seconds.

So there are two problems at once. People don't know what actually happens, and when you tell them, they don't believe it. A human expert joining your agent's session in under 30 seconds sounds made up if all you do is write it in a headline.

The site couldn't explain the handoff. It had to show it.

That's the reason there's so much animation on this site. It isn't decoration. The motion is doing the explaining, and every sequence is there because a sentence wasn't going to be enough.

I wireframed the whole site and got the founder's feedback on it before I opened a design file.

That mattered more than usual here because of the animation. I had two weeks. If a section moves after you've animated it, you're not adjusting a layout, you're redoing an After Effects file. Locking the structure and the order of the story early is what made it possible to spend three to five days on motion and still ship on time.

It made the feedback better too. When someone is looking at a wireframe they talk about what the section is saying. Once it's designed, they talk about colours.

Ten animations in total. Designed in Figma, animated in After Effects. Three to five days of the two weeks went into motion alone.

The hero was the important one. Instead of a static screenshot it plays the moment the product exists for. An agent is working, it gets stuck, an expert joins, the expert answers, and the agent carries on. You watch the handoff before you read anything about how it works, so by the time you get to the copy you already know what the product is.

The hardest one was the section under ONE PLATFORM. EVERY DOMAIN. We wanted to show the product actually working across different kinds of work, and doing that in a way that wasn't boring took the longest to figure out. Describing it in words is easy, and it's also forgettable.

What it became was a terminal. Real code runs, a Python traceback fails, the retries fail, and then the agent says it's requesting human help and an engineer connects. Anyone who has watched an agent loop on the same error knows that feeling straight away, and that recognition does more work than a paragraph would.

The rest of the sequences follow the same rule. If a section could show something instead of describing it, it shows it.

Humwork went from no website to a live one in two weeks, in time for the raise, with ten custom animations and the login and sign up screens included.