How Bytes AI Lifted Engineering Output 30% by Making Every Engineer's Weave Profile Visible

About Bytes AI

Bytes AI builds an AI phone assistant for restaurants. It answers every call, takes orders, handles reservations, and pushes orders straight into POS systems like Square, Clover, and Toast. It works in 15+ languages and remembers repeat customers. More than 1,000 restaurants run on it today.

Voice AI is hard to build. Latency, interruptions, noisy kitchens, and accents all have to work in real time. Bytes AI does this with five engineers, led by CTO Muthana Alhadrab.

The Challenge

Engineering changed. The way of measuring it did not.

Bytes AI first adopted Weave for a simple reason: visibility. Muthana and his management team reviewed most of the code themselves, but they had no metric tracking how output moved week over week or month over month. "We were just not seeing any actual trends," he says.

Then the models got good enough to change what engineering means at a startup.

"More than 95% of the problems we've had with our product, our development cycle, and literally everything else are solved with the latest coding models," Muthana says. Bytes AI is building complex, voice-AI-native products—hard problems—and still, the majority of those problems are solved with the latest intelligence.

If the model solves the problem, the bottleneck moves. Engineering becomes a question of how many problems you can push through the model at once. How many Claude sessions can you run in parallel? How many things can you work on at the same time? Claude will spend fifteen minutes, half an hour, an hour in a planning phase. How many other things can we do while that happens?

Muthana needed a way to see this.

The Solution

Reading Weave as time, attention, and effort

Muthana changed how he reads Weave. Engineering output, as measured by Weave, became his proxy for the time and effort invested in the work.

"If everybody is running four sessions for eight hours, everybody is going to have a very comparable outcome. The number of tokens you put into the problem, the amount of thought and precision you put into Claude, and how specific and difficult the problem is to diagnose and solve are all almost one-to-one with the metrics we see in Weave."

That simplified diagnosis. When output dips for a week, Muthana opens the timetable in Weave to see how long an engineer was active and looks for the pattern. Most of the time the answer is obvious before any investigation starts.

Muthana rolled Weave out to the team the same way he talks about it externally. At a startup, output is what matters. Between a mid-level and a senior engineer, the gap in output should be small when both are putting in the same amount of time with the same models.

At first, only Muthana and management could see the data. They used it to prep one-on-ones: pull up the report, see what someone worked on, see the improvement.

Then he opened it up. Every engineer signed in to Weave and could see their own profile, badges, and ranking against engineers on the platform.

The Results

30% more output in six weeks

Once everyone could see a finish line, behavior changed. One engineer started collecting Weave badges. Others watched their percentile. "When you log in to your profile and see that you are in the top 1% of all engineers for months on end, it is a pretty good feeling. It is a real accomplishment."

Six weeks after opening profiles to the whole team, engineering output was up about 30% across the board.

One founding engineer, Nader Abdulrub, has held a top-1% ranking on Weave for seven to eight months straight.

Why It Works

Bytes AI is a preview of what a small, AI-native engineering team looks like in practice:

• The models handle most of the problem solving. Engineers supervise, steer, and run many sessions at once.
• Output is measured, not guessed. Weave gives a shared, objective number for every engineer.
• The number is public inside the team. Visibility turned a management tool into a motivator.
• Leadership uses the same data to improve the environment: tooling, MCP servers, and workflows.