That’s simply how research labs are. Requirements pour in nonstop from the business units, from government affairs, from quality, and the managers spend their energy keeping up with them.
Meanwhile, the actual technology development tends to get pushed aside. My team is exactly like that. We’re the group that takes technology from advanced development into mass production, by combining parts that are already in production or still being developed. Three of us do the hands-on work; the other two are swamped with external communication. One person wears several hats at once — market monitoring, design paperwork, distributing spec sheets, certification. Of course frustration builds. For a long time, I used AI only as a tool that searched for things and polished my sentences. Then I changed my thinking. What if AI weren’t a tool, but an agent that could be a researcher’s hands and feet?
- In-house AI API
- Automating recurring, repetitive work · provided free inside the company
- Model
- Locked to an older version · latest not provided (cost policy) · we live with it
- Data export
- External transfer restricted · inside the corporate network only · security first
The key question was where to start. I picked work where three conditions overlapped: it comes around regularly, it’s tedious and time-consuming, and it isn’t very important. Tasks like that were scattered all through our process: monitoring market conditions, competitors, and industry news during advanced development; the paperwork that tags along whenever we run design software; writing and distributing product spec sheets; external communication; product certification. Examples of AI use had spread fairly widely across the company, mostly in data-focused departments, but none of them ran through the whole process, from advanced development to production development. We decided to fill that gap ourselves.
Not a tool that searches for you, but hands and feet that win back the development time a handful of researchers have lost.
Even the infrastructure was no easy matter. The company is strict about information security, so there are limits on sending data outside, and because of cost, the model isn’t the latest either. Still, we decided to be content that the company had opened up an API for internal use. Starting with what you already have in hand is how this notebook works. Honestly, it’s too early to talk about results. We’ve only just picked our first task to automate. How a small team can win back development time amid a flood of requirements, and how to work AI into the whole process, one step at a time — I plan to work through that over the next several issues. I don’t know yet how it will turn out. My approach comes down to one thing: with whatever we have now, even an in-house API on an older model, hand off first the work that comes around regularly, is tedious, and matters less, like market monitoring or distributing spec sheets. I’ll report back.
A researcher on a team that takes technology into mass production. Working where a handful of people weather a flood of requirements, I’m trying to set AI up not as a search tool but as hands and feet. I write more about what didn’t work than what did, and more about process than conclusions.