Selling call minutes instead of dials, and catching a live site that hadn't caught up

Today was a short one: 1 working session with coding agents, 4 prompts typed, about 2 hours of active work. It ended with 2 commits, 1 deploy to the website and 27 lines of code changed. That's a small diff for a change that affects how people understand what they're buying.

Dials out, call minutes in

The main change was to how the website describes the plans. They used to be framed around dials. Now they're sold in call minutes.

The longer I looked at "dials", the less it seemed to describe what the voice agent actually does. A dial is an attempt. It says nothing about the conversation that follows, or whether there is one. Call minutes are closer to the real work, which is time spent talking to someone. So I reworded the plans and shipped the change to the live site.

The live site didn't match

After my first round of changes, I checked the live site and it still showed the old dial wording. I'd assumed the change was out, and it wasn't.

That led to a fresh deploy, which got the new wording live. It also showed me a gap I hadn't noticed. I couldn't be sure the version of the website I was editing was the same one visitors were seeing.

So I recorded the site exactly as the live site serves it and committed that. Now our copy matches what visitors see. It wasn't exciting work, but I'd rather make changes on top of what's really live than on top of what I think is live.

A lot of agents at once

I also ran a large batch of helper agents in parallel, each on a separate area:

• site structure
• pricing
• the demo experience
• the demo voice
• compliance research

This is mostly research and groundwork rather than finished features. Running them in parallel meant I could cover several areas within the same 2 hours instead of going through them one by one. The hard part is the review afterwards. Each agent hands back something that looks finished, and I still have to decide what actually belongs on the website.

I also updated a test for the demo call flow as part of the day's changes. The demo is the first time most people will hear the voice agent, so I want that path covered by tests.

What I learned

• Check the live site, not your editor. I only noticed the stale wording because I looked at what visitors actually get.
• Keep the code matched to production. If they drift apart, every later change is built on a guess.
• Words in the plans are product decisions. Changing "dials" to "call minutes" was a handful of lines, but it changes what a buyer thinks they're paying for.
• Parallel agents move the bottleneck. They produce output fast, but judgement and review are still the slow part, and they're still my job.

A small day by the numbers. Still, the website now describes the product more honestly, and I trust that what I'm editing is what people see.