The state of AI coding tools in 2025
A lot has changed in the past two years.
In 2023, AI coding tools were a novelty. Developers tried them, were impressed by the demos, and mostly went back to their existing workflows. The suggestions were impressive in isolation but unreliable in practice.
In 2025, they're infrastructure. The question isn't whether to use an AI coding tool — it's which one, and how to integrate it into your workflow without creating new problems.
What's actually gotten better
Model quality has improved dramatically. The suggestions are better. The error detection is more reliable. The context windows are larger, which means tools can handle more of your codebase at once.
But the biggest improvement isn't in the models. It's in the integration. The best tools today feel native to the editor. They don't interrupt. They don't demand your attention. They're just there when you need them.
What's still broken
Context is still the hard problem. Most tools are still working with a fraction of your codebase. They don't understand your architecture. They don't know your conventions. They make suggestions that look right but aren't.
Latency is still an issue for some use cases. Real-time suggestions need to be fast — under 100ms ideally. Most tools are getting there for simple completions, but struggle with more complex suggestions.
Trust is the biggest unsolved problem. Developers are still not sure when to trust AI suggestions and when to be skeptical. Building that calibration takes time and good tooling.
Where this is going
The next frontier is agents. Not just tools that suggest code, but tools that can take a task, plan it, implement it, test it, and ship it. We're early, but it's coming.
The tools that win in that world will be the ones that developers actually trust. Trust takes time to build and seconds to lose. That's the constraint that shapes everything we do at Axon.