AI in 2026
My View on Where AI Is Heading
2024 and 2025 were years of exploration and experimentation for AI, with some already pushing into production.
In 2026, teams are no longer focused on what AI can do in theory. They are focused on whether it can be trusted, controlled, operated at scale, and justified by real outcomes. Reliability, cost, integration, and accountability are now central concerns.
Five forces stand out to me that will shape how AI is used, trusted, and integrated into real work.
1. AI Agents Proliferation
If they are not already, AI agents will be everywhere.
Sales, support, engineering, finance, operations. Every function will rely on agents, and most platforms will ship them out of the box. Creating an agent will feel routine, closer to spinning up a microservice than launching a new product.
We already see this direction with low-code/no-code tools like Microsoft Copilot, Salesforce Agentforce, and Vertex AI Agent Builder.
At the same time, with the mass adoption of coding agents like Claude Code and Codex, development itself is starting to feel faster, more iterative, and less constrained by team size.
But scale changes the problem.
As the number of agents grows, organizations will start to feel the friction:
- Agents are easy to create but hard to operate
- Chained autonomy compounds failures
- Debugging agent behavior becomes increasingly hard
In 2026, the question will shift.
It will no longer be “Can we create an agent?”
It will be “Can we trust the agents already running across the organization?”
2. Reliability, Trust, and Explainability
Now, to follow up, how do we build a trust and reliability layer when hundreds of thousands of agents are running across an organization, working alongside humans as virtual collaborators?
At this scale, reliability cannot be handled informally. Failures are no longer isolated, and small issues can cascade across systems. When something goes wrong, the hardest problem is often understanding what happened, why it happened, and who or what was involved.
Trust starts with visibility.
Similar to how air traffic control keeps planes coordinated and safe, organizations will need a 'command center' with real-time visibility into:
- Which agents are running
- What they are doing
- What decisions they are making
- How those decisions affect downstream systems
- Where errors or anomalies are occurring
Observability, traceability, and explainability become essential:
- End-to-end tracing across agent actions, tool calls, and handoffs
- Clear attribution of decisions and outcomes across humans and agents
- Human-readable explanations for critical decisions
- The ability to audit, replay, and reason about past executions
Explainability matters just as much as reliability.
As new agents, tools, and workflows are added, the system must remain understandable and governable. Organizations need architectures that allow agent fleets to grow without increasing operational complexity or risk.
Without this layer, agents may appear powerful but remain unreliable. Teams stop trusting systems they cannot see, explain, or evolve safely.
In 2026, trust in AI will not come from better prompts or smarter models.
It will come from systems that make agent behavior explainable, visible and controllable.
3. Agent Infrastructure Wars
As agents become easier to create and more widely available, the agents themselves will stop being the point of differentiation. Capability will converge.
What matters instead is the infrastructure that governs how agents run, interact, and evolve over time.
The real competition will move into the underlying layers that make large-scale agent systems usable:
- Orchestration layers that control execution order, coordination, and failure handling
- Memory systems that define what persists across time, sessions, and workflows
- Observability that explains why an agent acted the way it did
- Policy and safety enforcement that constrains behavior before things go wrong
Enterprises do not want black-box autonomy. They want control planes that let them shape behavior, limit blast radius, and intervene when needed.
This infrastructure must work across many agents, teams, and use cases. It must support versioning, rollout, rollback, and governance. It must integrate with existing systems, permissions, and compliance requirements. Most importantly, it must make agent behavior predictable enough to rely on.
The winners will not be the companies that ship the smartest agents.
They will be the ones that provide the most reliable, transparent, and scalable agent operating systems.
4. ROI Reckoning
The ROI conversation is coming to a head.
As AI usage spreads across organizations, spending becomes impossible to ignore. Token consumption grows, infrastructure bills rise, and latency starts to impact real workflows. At the same time, risk exposure increases, especially when hallucinations or incorrect actions affect customers, revenue, or compliance.
The systems that survive will be those designed with outcomes in mind:
- They tie directly to clear business KPIs rather than generic productivity gains
- They embed humans into decision points where judgment and accountability matter
- They limit autonomy and blast radius to control cost and risk
- They measure impact continuously, not just at launch
In this environment, AI will no longer be justified by promise or novelty.
It will be justified by results that can be observed, measured, and defended.
5. Tiny Teams or Tiny Individuals
Popularized by swyx, the idea of tiny teams captures a real shift already underway.
As AI agents and tooling improve, the leverage of individuals and small teams increases dramatically. What once required large groups can now be handled by a handful of people — or even a single person — coordinating AI systems alongside their own judgment.
This does not mean organizations will shrink overnight. It means the unit of execution is changing.
In 2026, impact will be less about headcount and more about:
- How well humans collaborate with AI systems
- How effectively work is decomposed and automated
- How much decision-making can be supported without removing accountability
Tiny teams move faster because coordination costs drop. Tiny individuals can own end-to-end outcomes because AI fills gaps across engineering, analysis, and operations.
But this leverage cuts both ways.
Without guardrails, oversight, and clear ownership, small teams can create outsized risk as easily as outsized value. As power concentrates, responsibility must as well.
In 2026, the advantage will not belong to the largest organizations.
It will belong to those that enable small teams and individuals to operate with leverage, clarity, and accountability.
What else I’m paying attention to
1. Multimodality and World Models
By the end of 2025, multimodality had improved significantly across the industry. Models became much better at understanding and combining images, video, audio, and text into a single flow of context.
We also saw early glimpses of world models in 2025, like Google’s Genie 3. These were early signals of systems that could reason about environments and outcomes, not just respond to prompts.
2. Generative UI
Generative UI explores interfaces that adapt in real time based on user intent and context, rather than relying on fixed screens and flows. Google’s GenTab experiments show how UI elements can be created, rearranged, or removed dynamically as tasks evolve. Over time, this could replace many static dashboards with interfaces that respond directly to what the user is trying to do.
3. Better Pre-Training and Post-Training Algorithms
Progress here is about making models behave better, not just bigger. Improvements in data quality, alignment techniques, fine-tuning methods, and evaluation loops lead to models that are more consistent, easier to control, and cheaper to run. These advances reduce unpredictable behavior and make AI systems more practical for production use.
4. Better AI Chips
Advances in AI hardware will focus on speed, efficiency, and cost. Lower latency enables real-time interactions, while better energy efficiency makes on-device AI more viable. As chips improve, more AI workloads can move closer to users and devices, reducing reliance on large centralized systems and unlocking new use cases that require fast, local decision-making.
Closing
The pace of AI progress in 2025 has been astonishing. Will we see the same trajectory in 2026? Who knows. 🤷
Either way, it’s shaping up to be an exciting year to watch. 😊