Great review, thanks for putting in that effort. I am using these coding tools in anger in developing our production apps at Fueld (www.fueld.ai). The last few weeks have been astounding. As background, I have been using the coding assistants as soon as the original co-pilot was released in VS Code, but moved to Cursor when it was released as it was AI first. Claude 3.5 and then 3.7 were my go to models. Gemini 2.5 Pro changed that and has been an amazing work horse in the last 2 months. I saw a big step in precision for very difficult tasks with o3, but it gets rapidly expensive and so I use a combination of Gemini for most tasks and o3 if it gets very tricky. The productivity is amazing. A tricky task involving data retrieval, aggregation and the developing of a complex UI to display it took me 2 days last year ... Gemini nailed in within 3 minutes yesterday ... I was astounded tbh and I am pushing these model's capabilities to the extreme daily. I felt even more humbled when the Gemini UI was even better than my own work last year. Claude 4 in production level coding is causing a few issues in Cursor already this evening, so it will take a little bit until those integration edges are smoothed out, but given the benchmark results, I am optimistic.
I appreciate hearing both what worked and what didn't. Balance over sensationalism makes your newsletter more valuable to me.
Great review, thanks for putting in that effort. I am using these coding tools in anger in developing our production apps at Fueld (www.fueld.ai). The last few weeks have been astounding. As background, I have been using the coding assistants as soon as the original co-pilot was released in VS Code, but moved to Cursor when it was released as it was AI first. Claude 3.5 and then 3.7 were my go to models. Gemini 2.5 Pro changed that and has been an amazing work horse in the last 2 months. I saw a big step in precision for very difficult tasks with o3, but it gets rapidly expensive and so I use a combination of Gemini for most tasks and o3 if it gets very tricky. The productivity is amazing. A tricky task involving data retrieval, aggregation and the developing of a complex UI to display it took me 2 days last year ... Gemini nailed in within 3 minutes yesterday ... I was astounded tbh and I am pushing these model's capabilities to the extreme daily. I felt even more humbled when the Gemini UI was even better than my own work last year. Claude 4 in production level coding is causing a few issues in Cursor already this evening, so it will take a little bit until those integration edges are smoothed out, but given the benchmark results, I am optimistic.