Unlocking AI-assisted development: Microdosing strategies for freethinkers
- Posted on March 19, 2024
- Estimated reading time 6 minutes
![AI assisted development strategies](/-/media/images/blogs/avanade-insights/ai-assisted-development-strategies.jpg?la=en&vs=2&w=1120&hash=BCF10BB5A7A9EBDB04A0B3C2E582726B)
If you read the tech press, it’s been hard to avoid stories about microdosing over the last year, from Wired to the Wall Street Journal, as people working in technology look for ways to boost creativity, mood and focus.
I’m not here to weigh in on that kind of microdosing, but it got me thinking: Just as some people look at microdosing as a thoughtful, measured way to explore something potentially scary but also profound and impactful (psychedelics, in this case), how could that approach help in other areas of work and life?
And that immediately got me thinking about AI-assisted software development — a growing source of anxiety, excitement and skepticism for many developers, sometimes all at once. Clearly, dev copilots are a big deal and here to stay, in one form or another. Thomas Dohmke, CEO of GitHub, has said that 80% of all code will be machine generated “sooner than later.” (Although in the same breath he says, “that doesn’t mean the developer is going to be replaced.”) But if you’re a developer, it’s enough to make you wonder how your work will change — and your job — no matter where you are in your career.
Both GitHub Copilot (GHCP) and AWS Code Whisperer claim to help developers code up to 55% faster, as my colleague Ben Leane related in an excellent post earlier this month. A six-month study of 450 developers at Accenture saw similar results — and more importantly, 90% felt more fulfilled with their jobs when using Copilot. Our own study at Avanade followed those trendlines: 81% of developers felt less frustrated and more fulfilled, and 96% said they were more productive and could focus on more satisfying work.
This is all promising, of course — but doesn’t that productivity boost mean that some jobs will be automated away? And how is this different from automation that we’ve seen before, qualitatively and quantitatively? We all know that with any disruptive technology, many jobs will change, some will go away and new ones will form. At Avanade, we take a view of informed optimism, in which AI augments humans, not replaces them, ushering in a more fulfilling future of work — but that big picture view isn’t much comfort when you’re worried about your individual job.
And here we get back to microdosing. If AI-assisted software development is here to stay, how can we engage with it in a thoughtful, measured way? As individuals, how do we make sure that if there is a tidal shift in the way that developers work (as many have opined), we’re riding that wave and not paddling to keep up?
I’m proposing four ways that you can “microdose” GitHub Copilot (or any similar copilot) — that is, introduce it into your work holistically and deliberately, in the spirit of adventure and experimentation. (And while I’m speaking to developers here, I hope it’s easy to imagine how this can apply to leadership in organizations as well.)
GHCP Microdose #1: Start incrementally
Bring GHCP into your workflow daily, whether via IDE or .com, or try OpenAI or Copilot for Azure. If you can’t use the client, use a reference architecture you’re familiar with. Start with A/B testing of your own, then try over and over to see what changes. The important thing here is to get hands-on with repeated use. Reading and watching can be valuable, but balance that with actual experimentation using your own projects. Use your judgment on whether to ask for help when you hit a block or to sit with an issue longer.
An important note here as you’re getting familiar with capabilities: Don’t file away how anything works as immutable, like “6 months ago I couldn’t do XYZ.” These tools are improving and changing monthly, so we need to keep learning as a community. (A side note on changes related to security: my colleague Josh McDonald has been keeping up with new risks.)
GHCP Microdose #2: Collaborate intentionally
Speaking of community, join one. In times of disruption and rapid change, we need to pool our experience, assets and even empathy. Rather than look for help only when you need it, seek out a collab that fits your mindset and interests, formal or otherwise.
GHCP Microdose #3: Implement broadly
Maybe it’s inevitable that most businesses will focus on improved productivity as the headline takeaway in the studies above, but that’s unfortunate and limiting. I think we’ll look back on this period and find that the biggest impact from copilots wasn’t incremental productivity bumps but unleashed creativity. What amazing things can happen when you release so many smart minds from so many tedious tasks? We’re about to find out.
To that end, I encourage you to try GHCP across your entire software development lifecycle — unit testing, data, initial code framing, code dev, maintenance, deployment, infra-as-code, everything. The barrier to experimentation and testing is dropping dramatically, so see where that can take you, whether that’s with modernization, greenfield, brownfield, different tech stacks, anything. Use GHCP to assess and augment your own maturity.
GHCP Microdose #4: Share enthusiastically
Many consume, few contribute. Many critique and theorize, but few bridge the gap in the early stages of these disruptive, powerful tech evolutions. This is your chance to not only ride the wave but to direct where it goes.
Hopefully the microdoses above will incrementally improve how you do development — removing toil and repetitive work, boosting speed, giving you more inputs when making decisions. But the next step is regularly sharing what you find with project leads, practice leads and global leadership in your organization.
Organizations are desperate for feedback to get telemetry and sentiment, and to understand what they actually need to do — down to every feature function, method step, QA process checkpoint, etc. As you master these evolving tools, you’ll have invaluable knowledge and insight into the methods and resources needed to harness them.
And that’s a good “microdose” to go out on! Let me know if you have success with any of these approaches. And if you have ideas of your own for small, deliberate ways to leverage GHCP or other copilots, share them with me and our readers below. We’re all in this together.
For more on software development with AI listen to our expert discussion on this podcast: Florin Rotar, Chief AI Officer of Avanade, Chris Lloyd-Jones, Head of Open Innovation at Avanade and Scott Hanselman from Microsoft discuss the future of AI-powered software development, the rise of GitHub Copilot X, and the responsible scaling of generative AI technologies within enterprises.
For more information on our application services visit Cloud and Application Services.
Comments