Predictions for 2025, 2030 and 2035
Predictions are hard, but they are necessary to check the quality of our mental models. I’ll check them again in 1 year.
2025:
The agency of AI agents will increase drastically.
They will have access to more tools and APIs.
"Test time" compute (or "reasoning") will evolve to search solution spaces more effectively. AI systems will test multiple ideas in parallel, advancing only the successful ones until finding a solution.
This approach will become mainstream in 2025, providing agents with significantly enhanced capabilities.
This approach will work using smaller and faster LLMs, allowing them to explore multiple solutions in parallel more quickly.
Kids who are motivated to learn will see their capabilities multiply tenfold.
Thanks to AI, almost every child has access to the most patient and knowledgeable teacher humanity has ever had.
Chatting with AI will become mainstream.
Mobile searches on Google will almost completely disappear.
People will use mobile apps less frequently.
The company with the best voice assistant will dominate the mobile market.
The concept of dev stack will embrace AI.
The usual stack was mostly based on some backend and frontend technologies.
AI will be mentioned in most dev stacks, explicitly listing tools used for AI inference, RAG (Retrieval-Augmented Generation), prompt monitoring, prompt engineering, and templating.
10x software engineers will become 100x software engineers. The market value of “not great” developers will decrease.
The productivity gap between average and exceptional engineers will widen dramatically. Poor developers will produce code equivalent to that of someone with no qualifications.
AI-first Development Teams Will Emerge.
An AI-first development team structures its entire workflow around AI tools and capabilities.
These teams will hire differently, prioritizing developers who demonstrate deep knowledge of AI tools and their effective use.
They will operate with remarkable efficiency—a team of 3 achieving what traditionally requires 20 people.
The core will consist of senior engineers with strong pre-AI experience who understand fundamentals, project organization, various tech stacks in detail, and software architecture.
Using the right tech stack will become even more critical.
With AI tools, choosing the right stack will create an even bigger competitive advantage.
Teams using the optimal stack will move 10 times faster than those using suboptimal technologies.
"One-person SaaS companies" will be a target segment.
The most innovative brands like Vercel will take this segment seriously and integrate it into their strategy.
30% of developers will spend more time chatting with AI than with humans.
Why ask your boss or colleague when an AI can answer you?
Remote workers will speak to their laptops.
We'll see a surge in remote workers using voice input. Having out-loud conversations with LLMs during work will become normal. In this setting, most writing will be done by voice.
The web will start being shaped for LLMs.
The same way Google was shaping the web indirecly by pushing every business to create SEO content, the new coming web will be shaped by the llms reading the content. They will need specific content created for them. This as started today.
The /llms.txt file – llms-txt
A proposal to standardise on using an /llms.txt file to provide information to help LLMs use a website at inference time.
In 2025 website owner will start design their website for LLMs, not only for humans.
API-first companies will grow faster than others.
Thanks to AI coding tools like Cursor, writing code has become much easier. As LLMs needs tools, New developers will increasingly rely on simple APIs to build most of their applications.
A web renaissance for micro websites.
All websites look nearly identical in 2024. With AI-powered coding tools, it becomes incredibly easy for creative individuals to build micro-websites that truly stand out.
I'm betting we'll see more and more websites like this one emerge everywhere.
LoRa will continue to be a pillar of modern Deep Learning Architecture.
LoRa will remain the de facto method for lightweight fine-tuning and task adaptation.
It is possible that LoRa will become a central component for agents learning new tasks.
2030:
Education has embraced AI.
Every education system is now structured around AI.
They have found clear answers about when to introduce it, when to authorize it, and when to forbid it.
80% of knowledge worker will spend more time chating with an AI than with humans during their working day.
Why ask your boss or colleague when an AI can answer you?
Automation thru AI will be mainstream.
Almost everyone will automate most of the boring things they do digitaly.
For the industry, automation will be the key for efficiency. Lots of roles will be created around AI automation. When you apply for a job in 2030, there's a good chance you'll be asked during the interview how you automated your job search process.
Knowledge management will become a central concept.
Knowledge management will become one of the key roles in every company. The unique knowledge that companies possess will serve as their primary competitive advantage.
All large enterprises will have a Head of AI.
Their role will be to identify automation opportunities across the organization.
They will collaborate with HR to manage large-scale organizational restructuring.
They will help strengthen knowledge management practices.
Most of enterprises are restructuring their dev team around AI.
Lean is the keyword.
Senior dev who embraced AI are in demand like never before.
One-person SaaS companies will become mainstream, with several reaching unicorn status.
This prediction isn't original—I'm simply endorsing it.
At least one country (or US state) will create a special legal status for "AI Agents."
While resembling AI citizenship, this will primarily serve as a loophole for enterprises to avoid liability for damages caused by their AI agents while retaining the profits these agents generate.
2035:
The world is still the same.
AI will unlock countless use cases that will transform how we work and interact with each other.
However, AI's current impact won't change society as dramatically as previous technological revolutions. Let's look at the big picture:
- The mastery of agriculture led to the creation of villages and towns.
- The tractor freed 95% of the population to pursue jobs beyond harvesting crops and tending to livestock.
- Computing and the internet enabled operations at scale, making possible today's global supply chain.
The most dramatic changes are behind us. In 10 years, even if some work roles disappear and others emerge, it won't fundamentally change life in modern countries when viewed from a macro perspective.
The essentials—how we're born, eat, travel, love, work (in front of a computer for many), and die—will remain largely unchanged.
While AI will accelerate the concentration of power, it will also enable counterbalancing forces to emerge. The implications of this power dynamic, however, are beyond the scope of these predictions.