If 2025 was the year artificial intelligence got a vibe check, 2026 will be the year the technology becomes practical. The focus has already shifted away from building ever larger language models and towards the harder work of using artificial intelligence. In practice, this includes developing smaller models where they fit, embedding intelligence into physical devices, and designing systems that integrate cleanly into human workflows.
TechCrunch experts spoke to see 2026 as a year of transition, one that evolves from brute-force scaling to researching new architectures, from flashy demonstrations to targeted deployments, and from agents that promise autonomy to those that actually augment the way people work.
The party isn’t over, but the industry is starting to rumble.
Scaling laws won’t cut it
In 2012, Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton’s ImageNet paper showed how artificial intelligence systems could “learn” to recognize objects in images by looking at millions of examples. The approach was computationally expensive, but made possible with GPUs. The result? A decade of hard-hitting AI research as scientists worked to invent new architectures for different tasks.
This culminated around 2020 when OpenAI released GPT-3, which showed how simply making the model 100 times larger unlocks abilities like coding and reasoning without requiring explicit training. This marked the transition to what Kian Katanforoosh, CEO and founder of artificial intelligence platform Workera, calls the “era of scaling”: a period defined by the belief that more computation, more data, and larger transformer models would inevitably drive the next major breakthroughs in artificial intelligence.
Today, many researchers believe that the field of artificial intelligence is beginning to exhaust the limits of scaling laws and will once again enter the era of research.
Yann LeCun, Meta’s former Chief AI Scientist, has long advocated an over-reliance on scaling and stressed the need to develop better architectures. And Sutskever said recently interview that current models are plateauing and training results have leveled off, suggesting a need for new ideas.
Techcrunch event
San Francisco
|
13-15 October 2026
“I think probably in the next five years, we’ll find a better architecture that will be a significant improvement on transformers,” Katanforoosh said. “And if we don’t, we can’t expect much improvement in the models.”
Sometimes less is more
Large language models are great at generalizing knowledge, but many experts say the next wave of business AI adoption will be driven by smaller, more flexible language models that can be fine-tuned for domain-specific solutions.
“Enhanced SLMs will be the big trend and become a key element used by mature AI enterprises in 2026, as cost and performance advantages will drive usage over out-of-the-box LLMs,” AT&T chief data officer Andy Markus told TechCrunch. “We’ve already seen enterprises increasingly rely on SLMs because, if configured correctly, they match larger, generalized models in accuracy for business applications and are excellent in terms of cost and speed.”
We’ve seen this argument before from French open-weight AI startup Mistral: It claims that its small models actually perform better than larger models on many benchmarks after detailing.
“The efficiency, cost-effectiveness and adaptability of SLMs make them ideal for custom applications where accuracy is paramount,” said Jon Knisley, AI strategist at ABBYY, an Austin-based business AI company.
While Markus believes SLMs will be key in the agent era, Knisley says the nature of small models means they’re better for deployment on local devices, “a trend accelerated by advances in edge computing.”
Learning through experience
People do not learn only through language. we learn by experiencing how the world works. But LLMs don’t really understand the world. they just predict the next word or idea. That’s why many researchers believe the next big leap will come from world models: artificial intelligence systems that learn how things move and interact in three-dimensional spaces so they can make predictions and take actions.
Signs that 2026 will be a big year for global models are multiplying. LeCun left Meta to create his own global modeling lab and is reportedly seeking a $5 billion valuation. Google’s DeepMind has branched out from Genie, and in August released its latest model that creates interactive, general-purpose global models in real time. Alongside demonstrations from startups like Decart and Odyssey, Fei-Fei Li’s World Labs launched its first commercial model globally, the Marble. New entrants like General Intuition in October raised a $134 million seed round to teach agents spatial reasoning, and video production startup Runway in December released its first global model, the GWM-1.
While researchers see long-term potential in robotics and autonomy, the short-term impact is likely to be seen first in video games. PitchBook predicts that the global model market in gaming could grow from $1.2 billion between 2022 and 2025 to $276 billion by 2030, due to the technology’s ability to create interactive worlds and more realistic characters without players.
Pim de Witte, founder of General Intuition, told TechCrunch that virtual environments may not only reshape gaming, but also become critical testing grounds for the next generation of foundational models.
Agent nation
Agents failed to live up to the hype in 2025, but a big reason for that is because it’s hard to connect them to the systems where the work actually happens. Without a way to access tools and environment, most agents were stuck in pilot workflows.
Anthropic’s Model Content Protocol (MCP), a “USB-C for AI” that allows AI agents to talk to external tools like databases, search engines, and APIs, proved the missing link and is quickly becoming the standard. OpenAI and Microsoft have publicly adopted MCP, and Anthropic recently donated it to the Linux Foundation’s new Agentic AI Foundation, which aims to help standardize open source agent tools. Google has also started deploying its own managed MCP servers to connect AI agents to its products and services.
With MCP reducing the friction of agents connecting to real systems, 2026 is likely to be the year agent workflows finally move from demos to everyday practice.
Rajeev Dham, partner at Sapphire Ventures, says these developments will lead to first-party solutions taking on “system of record roles” across industries.
“As voice agents handle more end-to-end tasks such as onboarding and communicating with customers, they will also begin to form the underlying core systems,” Dham said. “We will see this across sectors such as home services, technology and healthcare, as well as cross-functional functions such as sales, IT and support.”
Augmentation, not automation
While more hands-on workflows may raise concerns that layoffs may follow, Workera’s Katanforoosh isn’t so sure that’s the message: “2026 will be the year of the people,” he said.
In 2024, every AI company predicted that they would automate jobs without the need for humans. But the technology isn’t there yet, and in an unstable economy, that’s not really a popular rhetoric. Katanforoosh says that next year, we’ll realize that “AI hasn’t worked as autonomously as we thought,” and the conversation will focus more on how AI is used to augment human workflows, rather than replace them.
“And I think a lot of companies will start hiring,” he added, noting that he expects there will be new roles in governance, transparency, security and data management. “I’m quite bullish on unemployment averaging below 4% next year.”
“People want to be above the API, not below it, and I think 2026 is an important year for that,” added de Witte.
Getting physical
Advances in technologies such as small models, global models and state-of-the-art computing will enable more natural applications of machine learning, experts say.
“Natural AI will hit the market in 2026 as new categories of AI devices, including robotics, AVs, drones and wearables begin to enter the market,” Vikram Taneja, head of AT&T Ventures, told TechCrunch.
While autonomous vehicles and robotics are obvious use cases for natural AI that will undoubtedly continue to grow in 2026, the required training and development is still expensive. Wearables, on the other hand, provide a less expensive wedge with consumer buy-in. Smart glasses like the Ray-Ban Meta are starting to ship assistants that can answer questions about what you’re looking at, and new form factors like Health rings with artificial intelligence and smart watches they normalize the ever-active, on-body conclusion.
“Connectivity providers will work to optimize their network infrastructure to support this new wave of devices, and those with flexibility in how they can deliver connectivity will be in the best position,” Taneja said.
