Hint: don’t wait for the BEST tool
I have met developers with 8 different coding assistants installed in VS Code or traditional IDEs - they tell me that they use different coding assistants for very specific tasks that each tool is effective in. And the search for the ultimate coding assistant continues.
The best is yet to come? Link to heading
“Maybe the next frontier model with a highly customized interface will finally make me happy.”
But the truth is, the recent improvements from models and products are slowly stagnating. The improvement in capabilities from Claude 3.7 to Claude 4 can hardly be compared to the leap from GPT2 to GPT3 a few years back.
And we developers are a quirky bunch. Our love and pride for the mastery of eccentric and hard-to-use tools like Vim are fanatical. With still heated debates between spaces vs tabs, I doubt any coding assistant product will be able to conquer the hearts and minds of the majority developer market any time soon. Unlike VS Code, coding assistants are a lot more personal - personal in the suggestions and recommendations which also must fit into the personal styles and preferences of the developer experiences.
Last week, we learnt about the launch of Kiro, Amazon’s IDE built for the AI era of software development. There has been a lot of interest and demand for this new developer tool. I am loving the features - it provides much needed guidance to developers in their individual development processes.
Kiro shows great enhancements and has major improvements from Amazon Q Developer - a tool that I have worked with extensively with my customers as an AWS Software Architect.
But in the grand scheme of things, I am sure our tooling will continue to evolve, to be better.
The trap of local maximums Link to heading
Tools and technology have been changing at an incredible speed over the past few years and I expect them to continue to do so in the long run. We may discover a new fundamental architecture for LLM that is significantly more effective than Transformers. Who knows? Any major effort to optimize for the current toolkit will likely lead you to a local maximum, not a global one.
Given that the landscape of products and solutions will continue to evolve quickly, then, you should consider optimizing your engineering processes and culture with an AI-first approach - these changes will take more efforts and longer timeline but also more rewarding in the long run.
Coding assistants are primarily built for and used by developers. So investments in tooling tend to result in improvements limited to the developer persona. But the key to deliver innovation faster as an engineering organization is about scaling productivity gains across specialisations, teams and elevate entire organizations.
It is not uncommon these days to see engineering organizations sprinkling the “GenAI magic dust” all over their existing processes. “You are a developer, use AI to write your classes. You are a PM, use AI to write your user stories.” In these cases, the sum of the gain is not greater than the parts. At best, we are getting a % increase, not an X (5x better, 10x faster).
The processes and structures we followed were designed decades ago, for human engineering collaborations. This becomes the bottleneck when we adopt an entirely different class of technology today.
It’s like putting a jet engine on a horse carriage.
Scaling the gains of GenAI Link to heading
The coding assistants, models, MCPs, they all matter. We should continue to experiment with the latest toolings, but to also consider the other half of the solution.
Engineering leaders need to establish a way to scale what works well for individuals to the rest of the organization. Tools benefit individuals, processes help teams and culture elevate organizations.
And that is a difficult human mechanism to kick start, no AI agent can automatically get you there. But at the same time, this is also great flywheel and can accelerate businesses to be years ahead of the competition.