Hayden Cassar
Back to Blog
· 5 min read · Tools

AI and the Real Cost of Productivity: Beyond the 1000x Hype

A realistic look at how AI tools like Claude Code delivered a 1000x cost reduction - and the hidden investments that made it possible.

AI productivity development cost-reduction automation
AI and the Real Cost of Productivity: Beyond the 1000x Hype

AI reduced the cost of developing this feature by 1000x. From $11,450 to $11.45!

Are software development agencies doomed? Will clients vibe code everything they need now? Will agencies of 1000 developers shrink to agencies of 1?

Everyone currently loves writing AI hype posts.

They are self reinforcing. Also, no one wants to be “that guy” who didn’t think the internet would be a thing.

But, let’s get real for a second.

Did AI just actually reduce the cost of developing this feature by 1000x?

I could easily convince myself that it did.

$11.45 in Anthropic API credits consumed by Claude Code.

Ok, I created a PRD (Product Requirements Document) in a 30min screen capture, then did 2-3 hours of Code Review.

But still, very impressive as it’s by no means a simple feature (see video if you’re curious)

12 months ago I estimate that this would have taken 0.5 designers and 1 engineer 2-3 weeks. Let alone PM and QA. Easily $11,450.

A 1000x saving. It’s real.

But it misses a number of key points that make it less impressive;

  1. We just spent 9 weeks refactoring over 450,000 lines of Platfio source code to make it more accessible to AI.

  2. Platfio has a well established architecture; including CI/CD pipelines (which automate the deployment of code into production) previously developed at enormous expense.

  3. We built extensive tooling into our codebase specifically for AI agents. Including MCP servers, more tests, more documentation and extensive custom instructions; and

  4. We spent the last 18 months extensively training and upskilling our team in various tools including - Github Copiliot, Cursor, Codex and Claude Code.

Now lets really get real

  • Before we made the above 4 investments, Ai was only delivering trivial “tab to complete line of code” style improvements and was constantly breaking things when working agentically.

  • The 30mins of scoping a PRD and 2-3 hours of code review was extremely important. Without it the implementation of this feature using an agentic AI would have been a failure; and

  • It didn’t actually save 1000x because we would never have dedicated 0.5 designers and 1 engineer to this for 3 weeks to this feature. We have more important features in our backlog.

So… takeaways.

AI is not magic. It’s not as good as it seems. But it’s extremely impressive and getting better every month.

Turns out these classics are still true;

“If something sounds too good to be true it probably is”

And

“You get out of something, what you put into it”.

Clients will continue to need the help of agencies to build the things the need.