It's been a little over 2 months since I migrated Bondar Academy to a new platform I built myself. From scratch. Using Claude Code.
Previously, I was paying almost $300/month to LearnWorlds. Now it's around $35/month for my own LMS:
- Fly.io hosting $6/mo
- Supabase for database $25/mo
- CloudFlare for video storage and streaming for less than $3/mo
Overall, the "vibecoding" experience with Claude Code was quite mindblowing. In the first couple of weeks, I could not believe my eyes at how smart the agent was. I just described the feature I need, it builds it, and it works!
There is spaghetti code in the background, but for my goals, it does not matter. What matters is a working application with the features I need. And savings on the hosting :)
How is it related to testing, you may ask?
Well... DIRECTLY!
Testing is the final guardrail before the application goes to production. The user does not see the application code - they see design and behavior.
If the test is correct and it's passed, the application is safe to release.
Testing is a new bottleneck
Before the AI, the bottleneck of software development was always the coding and architecture. This was the slowest and the most expensive part of the process from feature idea to production release.
Now?
It's testing!
Writing the code is no longer the problem. But validating that the code is.
Here is a great video from Theo explaining how the software development life cycle is changing rapidly.
In other words, testing is becoming more valuable than ever. Developers can't do it well, so they need QAs. But running regression tests manually is no longer an option.
The speed of software development is so fast that regression testing MUST be automated. And it should be done quickly to keep up with the rapid delivery of new features.
Previously, we could afford to spend time inspecting the DOM, finding a suitable locator, building a Page Object architecture, or even worse, writing Cucumber step definitions that nobody ever read.
It's all in the past.
In today's world, you need to use a swarm of agents to write test automation code with Playwright or Cypress. Your job? Carefully review, make corrections, and refine the agent instructions as you go.
Instead of spending hours on perfecting the design of page objects, you need to spend time on refining agent rules and skills, so it can autonomously do the job for you with the highest success rate possible.
It is rarely be 100% every time, but it can do 90% of the heavy lifting for you, leaving the last 10% to your human expertise for the final tweaks.
The future of QA with AI adoption
So, where does this leave you as a QA?
The role isn't going anywhere. But it's quickly transforming!
Just being good at manually finding bugs isn't enough. Being good at understanding test automation frameworks is not enough either.
You should be good at leveraging AI as your skills multiplier!
Whatever you do well in your current role, you should create an agent with your custom instructions to 2-3x your productivity.
If you are on the market for QA roles, you'll get the job if you explain in the interview what agents you will create and how you will do it to speed up your testing.
You still need to thoroughly understand Playwright (or Cypress). You just don't write the code anymore.
You are a coding agent manager. You set the task and validate the result.
So I expect demand for this kind of engineer to be very high!
But guys... I didn't want to trigger a shot of anxiety in you :) Sorry if I already did.
Working with agents isn't that difficult and is actually really fun! I will show you soon!
Currently, I am working on the new course "Claude Code for Playwright", where I'll show how to use the Claude Code agent to write reliable Playwright tests fast.
It will be included in a Playwright UI Testing Mastery program. If you have already enrolled in this program, you'll get access automatically. I am planning to release this update sometime in May.
It's not the end, it's the beginning
Jevons Paradox - a 19th-century observation that when you make something more efficient, total consumption of it usually goes up, not down.
Between 1980 and 2010, around 400,000 ATMs were installed in the US, and bank tellers feared losing their jobs. But instead, bank teller employment grew from 500,000 to 600,000.
In 2016, Geoffrey Hinton – computer scientist and Turing Award winner – declared that "people should stop training radiologists now". AI can sometimes perform image recognition and reasoning better than humans. In 2025, American diagnostic radiology residency programs offered a record 1,208 positions across all radiology specialties, a four percent increase from 2024.
As software development becomes cheaper, more new software will be integrated into various parts of our lives. All this software should be tested.
So QA is not going anywhere; it's just transforming with AI, as roles in other industries have before. You just should stay open to those transformations.
Talk soon.
Artem
