InfoQ's AI, machine learning and data engineering team has published its 2026 trends report, naming AI agents the year's dominant theme and charting a shift from large language models to small language models. The report calls 2026 the year of inference, tracks the rise of physical AI and notes growing attention to AI evaluations. It was compiled by InfoQ editors with guest Chip Huyen.
Released on October 30 alongside a panel podcast, the annual report centres on an adoption graph based on Geoffrey Moore's Crossing the Chasm, which sorts technologies from innovators to early majority. The authors note it has been almost four years since ChatGPT launched in November 2022, and this year's edition adds new topics to its innovators category, promotes others to early adopters and removes technologies that have reached mainstream adoption.
Agentic AI leads the report's detailed findings. Agents are moving through several levels quickly, from personal agents helping one person to agents working with teams and companies using agents across entire organisations. Dominguez said the biggest change in 2026 is AI moving from something people interact with to something they delegate to, with systems that can plan, use tools, execute actions and keep working until a task is complete. Software development is probably where the transition is most visible, with coding agents that can navigate repositories, modify multiple files, execute tasks and iterate on their own.
On models, Casals said the model itself is becoming less of the whole story, as the question moves from which model is best to which model or combination of models offers the right capability, latency, data boundary and cost for a particular workload. He called the defining test of the moment not whether AI can do something once, but whether a team can trust it in a real workflow. The report also notes the rise of reasoning models that think through math and logic problems before responding.
Inference is expected to account for around two-thirds of AI compute spending by the end of the year, the report says, as reasoning models, long context and agents turn inference from a model-execution problem into a systems problem spanning compute, memory, networking, routing and economics. "The inference topic has exploded since the beginning of 2026," Luu said, pointing to open-weight models and stacks such as vLLM and SGLang that make it easier for companies, including financial firms handling personal data, to host their own inference services.
Physical AI, connecting software intelligence to machines that perceive and act in the physical world, saw significant developments in 2026, the report says, spanning robots, collaborative robots, autonomous vehicles and drones. Dominguez noted that unlike software agents, mistakes in the physical world carry physical consequences, making reliability, simulation, safety and verification more important. Huyen called robotics "the next frontier to get the models into the real world", while noting that a broken robot cannot simply be reset, which makes exploration far harder than in software.
The report also dubs 2026 the year of AI evaluations, with testing harness frameworks among the most discussed topics as organisations bring agents to production. New entries to the innovators category range from agent-to-agent protocols and harness engineering to spec-driven development, agent security and governance, and specialised decision models such as TypeSafeAI's Jev, described as running faster and at lower cost than LLMs. Looking to the next twelve months, the panelists predicted more innovation in US robotics companies, cheaper and faster models, limitations on AI in the areas of energy and AI infrastructure, and AI becoming an even bigger part of daily life, with niche areas settled where AI is genuinely valuable and where human intelligence is still required.
Analysis
Why This Matters
- The report maps where enterprise AI is heading over the next 12 months: agents moving from pilots to production, smaller, cheaper models, and compute spending shifting from training to running AI in production.
- For software teams, coding agents, spec-driven development and AI software factories point to developers spending less time writing code and more on requirements and specifications.
- The report's emphasis on evaluations and harness frameworks signals that proving agents do the right thing is becoming a condition of deployment.
Background
InfoQ is a publisher for software developers, and its AI, ML and data engineering team has produced annual trends reports and an adoption graph for several years, sorting technologies according to Geoffrey Moore's Crossing the Chasm categories of innovators, early adopters and early majority. This year's edition builds on the 2025 report, adding new topics at the innovators stage, reorganising overlapping categories and removing technologies that have reached mainstream adoption. The report arrives almost four years after ChatGPT's launch in November 2022 and two or three years after RAG technologies began seeing enterprise adoption.
Key Perspectives
Arthur Casals: The model itself is becoming less of the whole story; the defining test is not whether AI can do something once, but whether a team can trust it in a real workflow.
Daniel Dominguez: The biggest change in 2026 is AI moving from something people interact with to something they delegate to, and in physical AI, mistakes carry physical consequences.
Hien Luu: Inference has exploded since the beginning of 2026, and hosting open-weight models internally is becoming more practical; whether enterprises adopt fine-tuning at scale remains an open question.
Chip Huyen: The field is "going absolutely crazy", with investment climbing while exits lag, and robotics is "the next frontier to get the models into the real world".
What to Watch
- Whether agents move from pilots to production at scale: the report frames the coming year as the period when organisations work out how to operate, deploy and scale agentic systems.
- Energy and infrastructure limits: the panelists predict AI will face constraints in energy and infrastructure over the next year, a potential brake on growth.
- Enterprise take-up of fine-tuning and internal inference hosting, which Luu flagged as an open question heading into 2027.
Checked
12 claims in this story checked against 1 source: 12 supported.
Show each claim
- SupportedLuu said, “The inference topic has exploded since the beginning of 2026.”InfoQ: “Hien Luu : The inference topic has exploded since the beginning of 2026.”
- SupportedHuyen called robotics “the next frontier to get the models into the real world”.InfoQ: “Chip Huyen : It's the next frontier to get the models into the real world and we see a lot of progress.”
- SupportedThe report names AI agents as the dominant theme of 2026.InfoQ: “Similar to last year, Agentic AI has again been the dominant topic this year.”
- SupportedThe report charts a shift from large language models to small language models.InfoQ: “Key Takeaways Language models have shifted from large-sized models to the small language models.”
- SupportedThe report describes agents progressing from personal use to teams and organisations, and says organisations are working to deploy them to production.InfoQ: “Now the organizations are trying to figure out how to operate the agents, how to deploy them to production, and how to scale them in production.”
- SupportedDominguez said AI is moving from something people interact with to something they delegate to.InfoQ: “Daniel Dominguez : I think the biggest change in 2026 is that AI is moving from being something that we interact [with] to something that we delegate [to].”
- SupportedThe report calls 2026 the year of inference and expects inference to account for around two-thirds of AI compute spending by year-end.InfoQ: “2026 has been called the year of inference. The companies are shifting from training models to running AI inference in production. Inference is expected to account for around two-thirds of AI compute spending by the end of this year.”
- SupportedThe report calls 2026 the year of AI evaluations and says testing harness frameworks are among the most discussed topics.InfoQ: “The year 2026 has been dubbed as the year of AI Evaluations. There are a lot of different techniques that are gaining attention in this area. Testing harness frameworks are probably the most talked about topic this year.”
- SupportedThe report was released on October 30 alongside a panel podcast.InfoQ: “Oct 30, 2026 21 min read by Srini Penchikala Chip Huyen Daniel Dominguez Hien Luu Arthur Casals”
- SupportedThe report was compiled by InfoQ editors with guest Chip Huyen.InfoQ: “This report summarizes the InfoQ editorial team’s podcast with external guests to discuss the trends in AI and ML technologies for the year 2026”
- SupportedThe adoption graph is based on Geoffrey Moore’s Crossing the Chasm and uses categories including innovators, early adopters and early majority.InfoQ: “The categories are based on the book, Crossing the Chasm , by Geoffrey Moore.”
- SupportedThe report says physical AI had significant developments in 2026, including robotics, autonomous vehicles and drones, and notes that mistakes in the physical world have physical consequences.InfoQ: “Physical AI has seen significant developments in 2026. We are starting to connect the software intelligence to machines that can perceive and act in the physical world.”
