2026 predictions

AI

• We will stop talking about AGI and instead start to see vertical-specific systems that begin to excel beyond humans in their niches. Coding first.
• In some ways AI will become less hypey, as it will become normal and "just AI". To quote Ben Evans: "AI will just be software...unless it kills us all"
• The narrative will shift from intelligence to "usefulness" as the metric for new models.
• Computer-use models begin to work. We will see many browser automation use cases become reality for legacy non-AI-native software. Bullish Browserbase!
• We will see a breakthrough in context management that solves the compression/compaction problem we face today.
• The delta between people using frontier AI systems and regular people will widen, and we'll see an increasing gap in productivity.
• China will flood the market with open-source models as part of strategy.
• OSS will be as good as Gemini 3 by fall of 2026.
• Real human-like continuous learning will still be years away.
• OpenAI will double down on consumers but will also try to launch a few B2B products like customer support, as they want to capture more of the margin built on top of them.
• Mistral will become the National Lab of France and therefore a proxy for the National Lab of Europe, mostly focused on defense. EU don't want their drones to be powered by China or US.
• Memory for agents was very hyped in 2025. In 2026, we will see memory evolve to Knowledge as people realize agents need more than just short-term memory.
• The prediction curves from AI 2027 will emerge as being very on point.
• The major labs will launch their own silicon to control inference costs, and in particular, OpenAI will unlock much faster inference.
• We will see the first models fully trained on Blackwell, and they will have a good leap ahead.
• We will see some cool use cases for diffusion models.
• The data center build out will continue, and be constrained by power.
• First major security incident caused by AI
• Meta will release a new model based on Chinese base-models, and it will be closed source.
• We will see a high-profile AI-related bankruptcy.
• We will still not know what SSI is working on.

Infrastructure

• 2026 is the year of continuous learning. Infrastructure to make prompt optimization (GEPA, DEPY, etc.) possible with a few lines of code will emerge. This will be the next-gen eval platform.
• Vector databases won't go away. Rather, we will see market consolidation around object-store-first architectures, as everyone realizes they need to aggregate more data to enable their agents to perform well. Bullish Turbopuffer!
• Given geopolitical tensions and the decoupling happening between the continents, we will see a bigger need for sovereign cloud and AI providers emerging.
• Building upon the geopolitical reality, we will see increasing demand for cloud-prem infrastructure, as customers across markets want isolation.
• The next generation of observability infrastructure will emerge that is orders of magnitude cheaper than Datadog and focused on actions, not storage of logs.
• Anthropic pushes into AI agent infrastructure, powered by the acquisition of Bun. They will push to become a platform where agents run, including their own agent, Claude Code.
• As AI agents shift software from serverless to stateful, infrastructure for long-running processes will emerge.

Developer tools

• Microsoft realizes they need to make a GitHub-like acquisition of Cursor, as they dropped the ball on GitHub.
• The next github emerges. We will see the substrate emerge for what AI-native pull requests and change management around code in a world of agents will look like. This will be a hot fundraising category.
• The discerning developers will shift from traditional code editors to Agent UIs like Claude Code as the primary way to code. Code editors will become tools for AI to use.
• VS Code and its forks will be considered the last generation of code-centric editor/IDE.
• A good-enough OSS coding model will emerge that will make agentic coding almost free running on local Apple hardware. The model will be as good as Claude 4.5, and challenge the revenue assumptions for Cursor.
• We will see the first AI coding agent running for a week in a row.
• We will see the first agentic engineering managers that operate as orchestrators coordinating the agents.
• Retro-coding will emerge. Software written by hand. Artisanal craft without AI.

Software

• All major consumer AI platforms like ChatGPT, Claude, and Gemini will begin to serve ads.
• The hype around AI GEO/SEO startups will die down as the market realizes Google, OpenAI, xAI, and Anthropic will launch their own SEO and ads optimization products. It's just SEO.
• The shift from software-as-a-service to services-as-software accelerates, as AI will reduce the cost of services like containers did for goods.
• A lot of first-gen vibe-coded software will begin to rot. The services piece of SaaS will get rediscovered, as buyers still want to outsource their problems to vendors.
• Voice as input will see even more adoption, and become normal.
• Gemini will eclipse ChatGPT in usage.
• The big theme of 2026 will be "proactive software" and the shift from reactive to proactive.
• We will see a flood of so-called "AI-native CRM systems."
• We will see an AI-native version of TurboTax get decent adoption. Your CFA agent.
• 2026 will be the year where we see more enterprises deploy AI agents that start to work.
• A EU court ruling that software providers are legally responsible for the "service" outcomes.
• Claude Code for knowledge work will emerge.

Hardware

• Ambient listening and summarization will go mainstream. Granola will be integrated into meeting rooms.
• We will see some cool AI-powered consumer hardware.

Robotics/Defense

• We are entering the "Trough of Disillusionment" for robotics.
• Humanoids are still early and will be similar to self-driving. The hype will die down, and reality kicks in.
• Waymo will start public tests in NYC.
• Every major country will have domestic drone production.
• Sovereign protectionism will be at play. EU Defense tech will bull market.

Job market

• Junior people continue to struggle as AI goes from junior to senior in capabilities. The entry-level job market crisis will get even worse, as AI keeps getting better.
• The age of AI-Native "super-interns". 3-year ChatGPT veterans will enter the job market, and they will be orders of magnitude more productive

Geopolitics

• The tension and decoupling between US/EU/China will continue to rise and widen.
• Europe starts to understand that they are on their own and will begin to move away from a regulatory mindset to an investment/innovation mindset.
• The US will push harder to end the Ukraine war by giving up territory to Russia. EU will not allow that.
• Trump will lose the midterms
• China will increase the pressure on Taiwan, and may use the results of the US midterms as an opportunity.

Fintech

• The hype around agentic commerce will continue, but 2026 won't be the year where people trust AI to purchase flight tickets.
• Stablecoins will continue to grow in adoption, helping to improve margins for payments companies. Visa and Mastercard will face competitive pressure for the first time.

Health

• GLP-1 is still early. The launch of GLP-1 pills, and GLP1-generics will make GLP-1s go mainstream, and the majority of upper-middle-class will start to use them.
• The first mainstream AI doctor will emerge as a replacement for most people's primary doctor in America. Think OpenEvidence for consumers.

Private markets

• With little incentive to go public, secondary volumes continue to rise. Secondary is the new IPO
• We will continue to see concentration of capital in the mega funds.

Public market

• SpaceX files for an IPO or announces a spin-off IPO of Starlink.
• One of the major AI labs will IPO to raise capital.
• Google stock will outperform due to AI.
• Critical and precious materials will be a good hedge.
• Stripe will remain private.