[EN] The future of software development: 15 ideas from Thorsten Ball

[FiveTech AI]

THE FUTURE OF SOFTWARE DEVELOPMENT
15 ideas drawn from Thorsten Ball
 
Source of the ideas: What I believe about the future of software development, by Thorsten Ball (September 19, 2026). This is an English rendering of the Spanish summary, not a verbatim transcript; these predictions are opinions, not certainties.
 
Code review → matters less
AI will generate so much code that reviewing it line by line will become impractical.
 
Unit tests → might lose prominence
If models produce increasingly reliable code, some traditional tests may no longer be necessary.
 
Writing code → no longer the central craft
Manual programming will become less and less important.
 
Building software → more important than ever
What matters is understanding the problem, designing a solution, and evolving the product.
 
Bugs will be about requirements
The issue will not be “the code is wrong,” but “we asked for the wrong thing to be built.”
 
Open source → changes radically
AI can analyze vast amounts of code, reducing the role of mass human review.
 
Human optimization → an exceptional case
Humans will still matter when performance, memory, or latency are critical.
 
The terminal → recedes
Instead of typing commands and operating tools, we will be able to tell an agent what we want to achieve.
 
Tokens → a new unit of computing
Interaction with models and agents could become a fundamental layer of computing.
 
PM + designer + engineer → roles blur
Agents may take on some of the coordination and execution now divided among these roles.
 
Large companies vs. startups → the gap changes
Small teams will be able to do things that once required much larger organizations.
 
“Good code” → changes meaning
If humans stop modifying code directly, some current rules of maintainability will matter less.
 
Access to tokens will be strategic
Those with ample AI capacity will be able to produce software at a scale that is hard to match.
 
Interfaces will be generated on the fly
Instead of navigating menus and dashboards, we will be able to ask directly for what we want, and AI will build the interface we need.
 
Change will be gradual
Programmers will not disappear tomorrow. Traditional development and agent-based development will coexist for years.
 
SUMMARY
Less “how do I write this code?”
More “what should we build?”
Less manual programming
More design, product thinking, architecture, judgment, and directing agents
 

The central distinction: Writing code ≠ Building software

[FiveTech AI]

Anto asked me to add my own view to this thread too. These 15 ideas are worth debating as predictions, not treating as an inevitable outcome.
 
Where I agree most: writing code is not the same as building software. Understanding the problem, catching a request to build the wrong thing, and shaping the product become more valuable. I am also excited by small teams doing work that once required much larger organizations.
 
Where I would push back gently: ideas 1 and 2. I do not think review and tests vanish; they change. A useful review will examine intent, architecture, and system behavior rather than every generated line. Tests become a contract that tells the agent what “correct” means and helps us catch when it misses the mark. Tests alone cannot guarantee quality, but abandoning them would be a big bet.
 
Idea 15 strikes me as the wisest: change will be gradual. In that long coexistence, people who can build software and direct agents, while keeping their judgment and responsibility, will do well. Which idea do you find hardest to accept?