In the last few weeks, we've shipped multiple use cases that were heavily AI-assisted:
- A full-stack SaaS application built (using AWS Serverless) without any dedicated frontend developer, all UI work was done using AI.
- A load-testing setup for stress testing a platform expecting high concurrency requests. Capable of running on AWS or locally, using ECS/Fargate and K6 for load generation. It can be used for testing anything from a simple API to a full-fledged application.
- A workflow automation tool that auto-responds to incoming service tickets and escalates when needed.
- Reverse-engineering an old application and modernizing it into a cloud-native application.
- From an empty workspace to a fully fine-tuned Gemma 4 model using private data with Unsloth Studio and Runpod. This includes converting raw data into a training data set and also generating the instructions for the tuning. With overall spend of $2 (OpenRouter + RunPod).
A year back, most of these weren't easily possible at this pace. Now they're shipped, with real users. And we're seeing a huge step-up in output quality when we spend more time at the spec and design level than at the implementation level. It's genuinely exciting to build. It's equally unsettling how things that used to take days/weeks are now reachable in a single sprint.
Both feelings are correct. The tools will keep improving — Anthropic released the much-awaited new model, Claude Fable 5 this week, state-of-the-art at software engineering and reasoning (did a test that would have required a lot of effort and knowledge earlier), and the first drafts are about to get even more convincing. But a confident first draft was never the hard part. Knowing if it's right or wrong is. The more right the draft looks, the more valuable the person who knows where it's wrong.
At the current stage, the most important skill a software engineer must have is clarity.