For years my work ended with a Figma file and a handoff meeting. A few weeks later the build came back, and some of the small decisions were gone: the empty state, the error message, the spacing that made a long table easy to read. Nobody did it on purpose. A file just can’t carry everything.

Today I build what I design. I’m not a full time engineer, but the tools got good enough that a designer who knows some code can put a real product online. My two apps, Slovo and Recall, are live, and I use both every day.

What changed

Prototypes are real now. When I want to know if a flow works, I build it with real data in a day or two, and people click through the actual thing.

A design system lives in code. If the tokens and components only exist in Figma, the product drifts away from them within months, so I build them in both places at the same time.

AI does a lot of the typing. It drafts code, makes test data and helps me sort interview notes. I still decide what to build and how it should feel, and I check everything before it ships.

What stayed the same

The hard part was never the screens. It’s understanding who uses the product and what gets in their way. Someone checking hours on a phone on a building site in the rain needs something very different from an office manager doing invoices on a big screen.

Talking to those people, watching them work and writing the problem down before solving it takes as long as it always did. It matters even more now, because the build is fast and a wrong idea reaches users quickly too.

For teams that hire

When you look at a product designer’s portfolio this year, look for things that shipped. A designer who can own the flow, the design system and a working prototype saves a team weeks of back and forth. Engineers still build most of the product. They get fewer open questions and a better place to start from.

For me

Every project still starts with the same three things: users, goals, limits. I still sketch lots of rough options before I pick one. The last step is the part that changed. These days I hand over a link people can try, and the Figma file is one of the things behind it.

Most business problems aren’t new. Planning, invoices, customer lists and time sheets all have good existing tools. If a popular tool fits most of how you work, use it and adjust the rest of your process. It costs less than building, and you can start next week.

When existing tools stop fitting

Take a construction company with 40 people. The office plans in one tool, tracks hours in another, keeps materials in a spreadsheet and handles everything else by email. Every Friday someone copies numbers from one place to another. Nobody planned it this way. It grew over the years.

A few signs that you’ve outgrown existing tools:

• People copy the same data between systems every week
• You pay for five tools and use a small part of each one
• The way you work is part of why clients choose you, and the tools force you to change it
• People on site avoid the app because it was made for the office

Two options in between

Before building, two cheaper steps are worth a try. The first is no code: a tool like Airtable or Notion, connected to your other tools with Zapier or Make. It suits a small team whose process still changes every month.

The second is an AI agent. If most of the work is reading and replying, like sorting emails, taking numbers from invoices or answering the same customer questions, an agent on top of your current tools can take it off people’s hands. Someone still checks the result before it goes out.

When your own software makes sense

Build when your process is part of what makes you better than others, many people use the tool every day, it has to connect several systems and you plan to use it for years. For the construction company that could be one app with projects, crews, hours and materials in one place, and a simple phone view for people on site.

The first version doesn’t need all of that. Start with the part that wastes the most hours each week, use it for a month, then decide what comes next.

Check it in two minutes

I made a small tool for this: seven questions and one answer. Sometimes the answer is to buy, and that’s fine. Try Build or buy