Halftime Report: The Second-Half Predictions for AI in 2026

Prompts are becoming a smaller and less central part of how people interact with AI. The dominant interface is shifting from explicit requests to continuous, context-aware collaboration, where AI operates alongside users rather than waiting to be asked. In that model, prompting does not disappear, but it becomes occasional and situational, not the primary way work gets done.

The first half of 2026 has made that shift hard to ignore. AI is moving out of the prompt box and into the operating model. Enterprise use is no longer defined by one-off queries. OpenAI reports roughly 8× growth in weekly enterprise usage, 19× growth in structured workflows like Projects and Custom GPTs, and a 320× increase in reasoning-token consumption per organization. That pattern doesn’t just reflect more usage. It reflects AI becoming something closer to an embedded layer in the system—working continuously in the background rather than responding only when prompted.

Agents are only being used for a fraction of what they are capable of. Most users and organizations are still keeping them on a short leash. That’s about to change — and quickly. As their capabilities continue to advance, agents will begin taking on a much larger share of work, operating more like digital coworkers than tools, handling longer chains of activity, using systems, and coordinating across tasks in ways that weren’t possible even a year ago.

To date, most AI systems have remained in an advisory role. They draft, summarize, and suggest, then hand things back to a human. As agents become more capable and more persistent, the question becomes how much of the work they should be allowed to take on, and where people still need to stay involved. You can already see the early version of this shift. It’s no longer just about better models, but about systems that can operate across multiple steps and stay engaged over time. A consulting leader recently pointed to this directly, with concern moving away from incorrect outputs and toward unintended actions. Industry reports show teams beginning to move from single assistants to coordinated agents that can take on a much larger share of real work.

That doesn’t just expand capability. It changes how responsibility is distributed and how mistakes show up. When systems are involved in longer chains of activity, problems don’t always appear where they start. They show up later, often harder to trace, and sometimes after decisions have already been made.

The first half of 2026 has made that shift hard to ignore. AI is moving out of the prompt box and into the operating model. Enterprise use is no longer defined by one-off queries. OpenAI reports roughly 8× growth in weekly enterprise usage, 19× growth in structured workflows like Projects and Custom GPTs, and a 320× increase in reasoning-token consumption per organization. That pattern doesn’t just reflect more usage. It reflects AI becoming something closer to an embedded layer in the system—working continuously in the background rather than responding only when prompted.

Most companies are not going to hit a model or computing ceiling. They are going to hit a control ceiling, an inability to use their AI safely at scale. Governance is the next big bottleneck: clear oversight, usable audit trails, and rules that hold up once these systems start touching real work. That’s what is set to hold companies back more than anything else.

The gap here is already visible. Teams are placing AI inside workflows and decisions faster than they are building the systems to test outputs, monitor behavior, document changes, and intervene when something goes wrong. The more responsibility these systems take on, the less room there is for loose process.

As agents take on more responsibility and systems become more embedded, that gap stops being a governance issue in the abstract and becomes a practical limit on how far and how fast a company can actually go.

Policy is shifting from a stern suggestion to a market-defining force. It will affect how AI gets built, where it can be deployed, and which companies can use it first. Those that still treat policy as something to manage after the fact are going to find themselves boxed in by rules they did not help shape and are not prepared to meet.

You can already see this shift taking form. New state and federal proposals are pushing toward greater transparency, reporting, auditability, and formal oversight for advanced AI systems. The conversation has moved well past broad principles and into the mechanics of how these systems are built, tested, monitored, and disclosed. That matters because policy does not just create obligations. It starts shaping vendor choice, platform strategy, compliance cost, procurement standards, and how quickly a company can move once the stakes get higher.

We’re no longer in the App store age – tools, use-cases, one-offs are about to be a thing of the past. The next wave of advantage will come from redesigning work around AI, not from amassing a bigger toolkit. The companies that pull ahead will build systems: connected workflows, defined roles, and operating rules that let AI support real execution across the business.

That shift has already started. The biggest gains are coming from putting AI inside structured workflows and reworking how tasks move between systems and teams. In practice, that means changing process design, handoffs, approvals, and ownership, not just adding a new model on top of old work.

This is where the earlier signals connect. Agents take on more work. Governance sets the limits. Trust determines where adoption holds. The companies that align those pieces will move faster and with fewer self-inflicted problems.

The companies that succeed in 2026 and beyond won’t be because of their budget or which AI models they chose to use. It will come down to how clearly they understand what they are trying to achieve and how well they’ve built their business around it.

AI is showing up across more parts of the organization, but that doesn’t determine what it improves. The outcome still depends on the underlying growth strategy and how the work, AI, and everything else – supports it. Where that direction is clear, AI tends to reinforce it and help it move faster. Where it isn’t, it usually leads to more activity without much change underneath.

If you’re thinking about how well your AI efforts connect to your growth strategy, or whether you have one at all, it’s worth taking a closer look now. The right alignment tends to show up quickly in results – and just as quickly when it’s missing.