AI-assisted testing, code review, validation, CI failure fixing, and UI regression detection

I’ve been using assisted AI for coding since early 2023, but 2026 is when everything changed and I started using it full-time on Readplace. At the time of writing this, readplace.com has more than 3000 commits.

So yes I fully jumped on the AI bandwagon.

I'm not a software engineer.

Today I taught a room full of them.

The topic was AI-driven development. Not a demo. The actual practice: how you plan the work when an agent writes the first pass, what you still review line by line, where a human has to stay in the loop, and what changes when writing the code stops being the slow part.

I walked in thinking I was the least qualified person in the room. On raw coding, I was. These folks have shipped more production code than I will ever read.

That ended up being the useful part.

Two things I took from it:

1. AI-driven development is a practice, not a title. It rewards people who think in process, systems, and guardrails. That is my day job. The syntax was never the hard part.

2. Expertise didn't go away, it moved. Nobody in that room got replaced by the tooling. They became the reviewers and the architects, the ones who catch what the model gets confidently wrong.

I left more impressed by them, not less. And proud to have had the seat.
If you're waiting on the right title before you say something about AI at your company, I think you're waiting on the wrong thing.

Curious what you all think. Is AI closing the gap between technical and non-technical roles, or just moving it somewhere else?

#AI #Automation #AIatWork #FutureOfWork #GenerativeAI

The way we use AI while building a feature has changed quite a bit.

Earlier, we mostly thought of AI as something that helps us write code faster.

Now, we try to bring it into the process much earlier.

For a new feature, we first give AI the context it needs — relevant code, documentation, architecture, and our project-specific skills/guidelines.

Then we use Plan Mode and ask:

“Understand the existing implementation and documentation. Identify dependencies, edge cases, assumptions, and missing information. Ask all necessary questions and create an implementation plan. Don’t change the code yet.”

We review the plan, make adjustments, and only then ask it to implement.

After implementation, we use AI again for a different job:

“Review this like a senior engineer. Look for incorrect assumptions, race conditions, failure scenarios, performance issues, and missing tests.”

The interesting part for me is that AI becomes much more useful when we give it a way of working, not just a task.

Skills tell it how we build things.

Documentation tells it how our system works.

Plan Mode helps it reason before changing things.

So instead of:

“Write this feature.”

It becomes:

Understand → Question → Plan → Build → Review

That shift has made AI feel much more like an engineering partner than just a coding assistant.

What has worked best for you when using AI for development?

#AI #SoftwareEngineering #DeveloperProductivity #ArtificialIntelligence #Coding #SystemDesign #Engineering #OneThingToKnow

Go beyond chat in the GitHub Copilot app with these slash commands. They’ll help you plan, collaborate, automate, and customize your dev workflow.

✅3 simple ways on how I use AI as a software engineer at Google in my day to day life.

1️⃣. Understand Faster

Whenever I pick up a new feature or work area, I ask AI to explain the code flow, tell me the services involved and generate sequence diagram.

The main thing I ask for is the exact files and code links of the flow. Then I verify every step, it’s still some manual effort but I am sure of my understanding, I gain the knowledge for future and am 100% confident to make any design choices.

2️⃣. Code Better & Faster

AI is great at writing code. But it might skip some details, like sometimes a function may already exist in the hierarchy but it defines it again.

I usually avoid it with the first step- understanding things in and out. I also prompt AI to review the code as a lead engineer.

The same goes for tests, there can be issues. I always make sure to list down all the edge cases I can think of. Then I compare that checklist against the generated tests to make sure nothing is missing. I once wrote 45 test cases for a 8 line change.

3️⃣. Write Docs Faster

Take design doc for example: I attach all reference docs where I got all the info from & a design template & then I turn on voice mode and just talk and explain my thought process.

I get a really good draft and starting point. From there, I iterate and if it’s not working out in 5 prompts I write some sections manually(great for long term memory).

For me, AI really helps me with everything, has made me so much productive and saves me a lot of time.

There are many other ways I use AI, for example custom agents etc. Let me know in the comments if I should share that next.

#AI #SimpleUseCases #Productivity

Today I've been putting Opus 5 head to head with Sol and honestly the difference is stark. So I started out on just the $20 plan and the first thing I noticed was that running Opus 5 on Extra reasoning was burning through my weekly limit about as fast as Sol on High did on my 5X Codex Pro plan (20$ vs 100$, lol). As for the actual comparison: For planning Opus 5 is way more goal-oriented. No fluff, just a straight coherent plan and it actually keeps you involved the whole time you're designing it, which for me is a good thing. After using Sol since release my AI workflow and project rules had gotten so complicated I'd basically lost track, I didn't really know where things stood anymore. Opus brought me back into the project, now I actually know what's going on. Sol on the other hand just overengineers everything, runs in circles, and loses me completely along the way and unfortunately it produces code I can barely look at without needing to go breathe some fresh air.

To be fair Sol is still a genuinely strong model, it's just better as a reviewer than a planner. There it's excellent.

Opus 5 walks you through its reasoning and stays concrete instead of drifting into pseudo language. Like, I kept catching myself typing "ELI5" at Sol after a long answer, and I'd like to think I'm not that dumb. With Opus I actually read the whole answer and understood it.

So yeah, Sol produces these plans that sound impressively sophisticated, good enough to break into the Pentagon, but by the time it gets there you've blown through your weekly limit.

I have upgraded to 5X CC and downgraded to 20$ Codex so I can still use Luna / Terra for small, well-defined implementation packets.

Rider 2026.2 opens up the IDE’s own intelligence to your AI coding agents, so they work from real project knowledge instead of reconstructing it from files and terminal output.

A new set of agent skills covers testing, profiling, refactoring, and official Microsoft .NET workflows, and GitHub Copilot now joins the lineup as a natively integrated agent. This release also delivers a wave of performance gains that make the IDE feel faster across both .NET and game development.

Why dynamic workflows exist, the three failure modes they fix, the six patterns, and how the harness runs. Beginner-friendly.

AI didn't make me a 10x developer

It removed hours of boring work.

Now I use it to:
• Design APIs
• Generate boilerplate
• Review code
• Find edge cases
• Write tests
• Create documentation

The biggest productivity gain isn't coding.

It's reducing context switching.

How are you using AI in development?

Let me save you some time upfront: I’m not going to tell you it changed my life.

A month ago, I wasn’t even looking for a new AI coding tool. I already had a workflow that worked. Claude.ai for code reviews, Axios for API calls, VS Code for everything else. Then I kept seeing developers talk about Claude Code — the terminal-based version, the one that reads your entire codebase automatically.

Some said it was the best coding assistant available. Others said it wasn’t worth the price. Most reviews were either suspiciously positive or completely dismissive.

So I stopped reading opinions and started testing.

For 30 days, I used Claude Code on a real production project. Not a side project. Not a tutorial app. A form-heavy medical examination system built for a state government — complex conditional logic, multiple examination form components, Java DTO integrations, file upload flows, and zero tolerance for bugs.