If 2025 was the year that confirmed the vibe of AI, 2026 will be the year that the technology becomes commercially available. The focus is already shifting from building ever-larger language models to the more difficult task of enabling AI. In practice, it involves deploying smaller models in the right places, embedding intelligence into physical devices, and designing systems that integrate cleanly into human workflows.
Experts TechCrunch spoke to said 2026 will be a year of transformation, from aggressive scaling to exploring new architectures, from flashy demos to targeted deployments, and from agents promising autonomy to agents that actually extend the way people work.
The party isn’t over, but the industry is starting to calm down.
Cannot be solved by scaling method
In 2012, Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton’s AlexNet paper showed how an AI system can “learn” to recognize objects in images by looking at millions of examples. Although this approach was computationally expensive, it was made possible by the use of GPUs. result? A decade of serious AI research where scientists worked to invent new architectures for a variety of tasks.
This situation reached its climax around 2020 when OpenAI announced GPT-3. This showed how simply making the model 100x larger can unlock capabilities such as coding and inference without the need for explicit training. This marks a transition into what Kian Katanforoosh, CEO and founder of AI agent platform Workera, calls the “era of scaling,” defined by the belief that more compute, more data, and bigger transformer models will inevitably drive the next big breakthrough in AI.
Many researchers now believe that the AI industry is starting to exhaust the limits of the laws of scaling and will once again enter the era of research.
Yann LeCun, former chief AI scientist at Meta, has long argued against over-reliance on scaling and emphasized the need to develop better architectures. Sutskever also said in a recent interview that current models have plateaued and pre-training results have plateaued, indicating the need for new ideas.
tech crunch event
san francisco
|
October 13-15, 2026
“I think in probably the next five years we will find better architectures that will significantly improve transformers,” Katanforoosh said. “If we don’t do that, we won’t expect much improvement in the model.”
Sometimes less is more
While large language models are great at generalizing knowledge, many experts say the next wave of enterprise AI adoption will be driven by smaller, more agile language models that can be fine-tuned for domain-specific solutions.
“Fine-tuned SLMs will become a big trend and a staple used by mature AI companies in 2026, with cost and performance advantages that will drive their use over out-of-the-box LLMs,” Andy Markus, chief data officer at AT&T, told TechCrunch. “We are already seeing enterprises increasingly relying on SLM because, when properly tuned, SLM can match larger, generalized models in accuracy for enterprise business applications and outperform in terms of cost and speed.”
We’ve seen this discussion before from French openweight AI startup Mistral. The company claims that its smaller model actually performs better than its larger model on several benchmarks after tweaking.
“The AI strategist at ABBYY, an Austin-based enterprise AI company, says:
While Markus believes SLM will be key to the agent era, Knisley says the nature of small models means they are well-suited to deployment on local devices, and “advances in edge computing are accelerating this trend.”
learn through experience
Humans don’t just learn through language. We learn how the world works by experiencing it. But LLMs don’t really understand the world. Just predict the next word or idea. That’s why many researchers believe the next big leap forward will come from world models, AI systems that can learn how things move and interact in 3D space to make predictions and take action.
There are increasing signs that 2026 will be a significant year for global models. Mr. LeCun is leaving Meta to start his own Global Model Lab and is reportedly seeking a $5 billion valuation. Google’s DeepMind works on Genie and in August announced its latest model for building real-time, interactive, general-purpose world models. Alongside demos from startups like Decart and Odyssey, Fei-Fei Li’s World Labs announced Marble, its first commercial world model. In October, startups like General Intuition secured a $134 million seed round to teach spatial reasoning to agents, and video generation startup Runway released its first global model, GWM-1, in December.
While researchers see long-term potential for robotics and autonomy, short-term effects may first be seen in video games. PitchBook predicts that the market for world models in games could grow from $1.2 billion between 2022 and 2025 to $276 billion by 2030, driven by technology’s ability to generate interactive worlds and more lifelike non-player characters.
General Intuition founder Pim de Witte told TechCrunch that virtual environments could not only be used to reimagine games, but also be an important testing ground for next-generation foundational models.
proxy state
Agents failed to live up to the hype for 2025, and a big reason for that is the difficulty of connecting them to the systems where the work will actually be done. Without a way to access tools or context, most agents were stuck in the pilot workflow.
Anthropic’s Model Context Protocol (MCP), the “USB-C for AI” that allows AI agents to communicate with external tools such as databases, search engines, and APIs, has proven missing connective tissue and is rapidly becoming the standard. OpenAI and Microsoft have publicly embraced MCP, and Anthropic recently donated MCP to the Linux Foundation’s new Agentic AI Foundation, which aims to help standardize open source agent tools. Google is also starting to launch its own managed MCP servers to connect AI agents to its products and services.
Because MCP eases the burden of connecting agents to real systems, 2026 could be the year that agent workflows finally move from demo to everyday practice.
Rajeev Dham, a partner at Sapphire Ventures, says these advances will allow agent-first solutions to take on the role of “system of record” across industries.
“Voice agents will begin to handle more end-to-end tasks, such as reception and customer communication, and will also begin to form the underlying core systems,” Dahm says. “This will be seen not only in horizontal departments such as sales, IT and support, but also in various sectors such as home services, proptech and healthcare.”
Augment, not automate
While more agent-like workflows may raise concerns that there will be layoffs, Workera’s Katanforoosh isn’t sure that’s the message.
“2026 will be the year of humanity,” he said.
He predicted that by 2024, all AI companies will automate jobs without the need for humans. But that’s not common rhetoric at a time when the technology is not yet widespread and the economy is unstable. Over the next year, we will realize that “AI is not working as autonomously as we thought,” Catanforouch said, and the conversation will focus on how AI is being used to augment human workflows, rather than replace them.
“And I think a lot of companies will start hiring,” he added, saying he expects new roles to emerge in AI governance, transparency, safety and data management. “I’m pretty bullish that the unemployment rate will average below 4% next year.”
“People want to be above API, not below API, and I think 2026 will be a key year for this,” De Witte added.
train your body
Advances in technology such as small models, world models and edge computing will enable more physical applications of machine learning, experts say.
“Physical AI will become mainstream in 2026 as new categories of AI-powered devices such as robotics, AVs, drones, and wearables begin to enter the market,” Vikram Taneja, head of AT&T Ventures, told TechCrunch.
Self-driving cars and robotics are obvious use cases for physical AI that will no doubt continue to grow in 2026, but the necessary training and deployment will still be expensive. Wearables, on the other hand, offer a cheaper wedge with consumer buy-in. Smart glasses like Meta’s Ray Ban are starting to ship assistants that answer questions about what they’re looking at, and new form factors like AI-powered health rings and smart watches are making always-on, body-worn reasoning the norm.
“Connectivity providers will look to optimize their network infrastructure to support this new wave of devices, and those who are flexible in how they deliver connectivity will be best positioned,” said Taneja.
