Five speakers identify the biggest mistakes companies make adopting AI: layering tools onto unchanged workflows, moving too slowly with token budgets, running scattered pilots instead of focusing on one high-impact use case, and cutting headcount before fixing technical debt. These missteps block real productivity gains.
What is the biggest mistakes that companies are making right now with AI adoption?
I think the primary mistake is companies are trying to bolster the AI tools into their current process without changing the process. Previously, coding was the bottleneck. Writing code out was the longest pole and all the process was built around that. Now it has shrunk. So all the other bottlenecks like code review, CI/CD, CD pipeline. They are coming in the way. So if you don't change that process, you're not going to get the benefit out of the AI tools.
I think a lot of companies are being way too slow with their adoption. I think we're in the phase where you need to give your engineers a lot of tokens and allow them to figure out the best ways to use these because the technology is changing so, so quickly. So trying to wait, I think, is a bad strategy.
The biggest mistake they're making is that they are just there's this proliferation of AI pilots and different initiatives happening, right? And, you know, they're not focused on a single area where they can improve themselves, you know, and deliver measurable impact, you know. But, you know, they're just wasting money, bleeding resources, you know, on a million different initiatives that are leading nowhere, basically.
The biggest mistake with AI adoption is immediately using it as a cost savings measure. So if you get a little bit of productivity growth, what some companies do, right, is immediately reduce headcount rather than dealing with technical debt and building for the future.
There is a time and place for AI, and right now everyone is moving fast and using the large language models for just about every task that needs to be done, which is totally fine for getting started. But there is a right tool for every job, and it's not always Claude or Gemini or ChatGPT to do that. Like, we typically say it's like cracking a walnut with a sledgehammer. It will do the job. Maybe it's not the right tool for the job. So I think the biggest mistake companies are making is really using these large language models for tools that we know much smaller models can do the same task much more efficiently and much cheaper.
