AI's Second Moment: When Agents Go From Hype to Reality

Something shifted over the holidays. MidJourney CEO David Holz dropped a telling observation: "I've done more personal coding projects over the Christmas break than in the previous 10 years combined." It wasn't just him. While people were away from their desks, a new generation of AI tools was quietly rewriting what felt possible.

The AI Daily Brief's Q2 2026 report calls this "AI's second moment." If the first moment was ChatGPT proving chatbots could be useful, the second is about agentic systems—AI that doesn't just assist but executes. The report synthesizes data from 400+ sources and monthly user surveys to map a quarter where AI moved from speculative promise to operational reality, with all the disruption and dysfunction that entails.

The Numbers That Changed The Conversation

The economic stakes escalated dramatically. Hyperscalers are planning $650 billion in capital expenditures this year—three times their spending from two years ago, exceeding the inflation-adjusted cost of the U.S. interstate highway system. Not venture-scale bets anymore. Infrastructure-scale commitments.

Claude Code, Anthropic's flagship product, jumped from $1 billion to $2.5 billion in annualized revenue in two months. Cursor doubled from $1 billion to $2 billion. Lovable hit $400 million ARR with a $100 million spike in a single month. Anthropic now commands 70% of first-time enterprise AI buyers, according to Ramp statistics, with OpenAI at 25%.

The market narrative flipped on its head. Q4 2025 was all about the "AI bubble"—what if this doesn't work? Q1 2026 became about the "SaaS apocalypse"—what if it works too well? Public software companies saw widespread carnage. Block cut 40% of its workforce. Investor concern shifted from "what if AI plateaus" to "what if AI eliminates entire categories."

The Agent Explosion

At the center of this shift sits OpenClaw, which the report calls potentially "the most important software release ever," quoting NVIDIA CEO Jensen Huang. Starting as Claudebot in January, the project evolved rapidly and became the most-starred open source project on GitHub. Within weeks, the industry was racing to integrate similar capabilities.

Notably, Anthropic and OpenAI are converging from opposite directions. OpenAI started with product sprawl—multiple standalone tools—and is consolidating inward into a super app. Anthropic began with a single dominant product and is expanding outward to make it extensible. Different strategies, similar destination.

The practical deployment numbers tell the real story. Gartner projects 40% of enterprises will have working agents in production by end of 2026. Ramp and Stripe have launched agent credit cards—AI that can actually spend money. NVIDIA's Nemo Claw represents what's coming: enterprise-grade hardening of agentic tools for actual business use.

One data point captures the extremes: Pulsia, a company that produces "fully agentic companies," reached $6 million in annualized revenue with a single founder and zero employees. Whether that's durable or novelty revenue remains open. But as founder Ben Sarah noted, "the zero employee company isn't a thought experiment anymore. It's a live dashboard with weekly metrics."

The Capability Overhang

Here's where it gets uncomfortable. The report's user surveys show a widening gap between what AI can do and what organizations are actually deploying. In legal, Anthropic research found AI capable of handling 80% of certain tasks, but only 15% showed actual adoption. Finance firms report 91% are using AI but also 91% see fairly low impact, citing data quality as the primary barrier.

The type of value is shifting too. In late 2025, time savings dominated AI ROI. By February 2026, time-saving use cases dropped from 19.9% to 13.6% of reported value. Increased output and throughput became number one. New capabilities jumped from 22% to over 26%. This is the transition from efficiency AI to opportunity AI—not doing the same work faster, but doing different work entirely.

Barriers remain consistent: time, policy, and skill gaps. But the cost of the capability overhang is rising. The gap between AI leaders and laggards is expanding, and the competitive disadvantage of being slow grows with each model release.

When AI Got Political

The quarter's most volatile development was Anthropic's collision with the Pentagon. Reports emerged that Claude had been used during operations against Venezuelan President Nicolás Maduro. Anthropic wanted commitments against autonomous weapons and citizen surveillance. The Pentagon wanted "all lawful use" rights.

Defense Secretary Pete Hegseth issued ultimatums. Anthropic refused to comply. They became the first U.S. company designated a supply chain risk by the Department of Defense. Anthropic sued. Claude continued being used in the Iran conflict. The situation remains unresolved.

When OpenAI announced they'd signed an agreement with the Department of Defense the same night as Anthropic's ultimatum deadline, backlash was swift. ChatGPT saw a 775% surge in one-star reviews. Claude hit number one in the app store for the first time.

Meanwhile, data center politics emerged as a parallel front. State and congressional campaigns focused on infrastructure costs led to commitments from hyperscalers that Americans wouldn't foot the bill for the AI buildout through higher electricity costs. The "anti-AI movement"—really a collection of different grievances—made Time magazine's cover with "People vs. AI."

The White House released its AI legislative framework toward quarter's end, which the report frames as "an opening salvo in a heightened stakes conversation around AI policy."

What The Users Actually See

Monthly user surveys from AI Daily Brief show practitioners are already living in the agentic future. 71% have "vibe coded" in the past month—using AI for coding without traditional programming. 62% report automation or agentic use cases rather than just assistance. The average respondent uses 3.5 different models, taking a portfolio approach and choosing the best tool for each task.

No single model dominates across benchmarks anymore. GPT, Claude, and Gemini trade positions depending on the test. Users are becoming model-agnostic, optimizing for outcome rather than platform loyalty.

Customer service shows 91% of businesses experimenting with AI chatbots, though 64% of customers prefer no AI in their interactions—a preference that's probably not coming back. HR saw 320% growth in AI deployment over 12 months, from 19% to 61%. Sales leads in maturity, with 63% of tracked use cases in the "prime time" category, ready for most organizations.

Marketing is birthing entirely new categories. Generative Engine Optimization (GEO)—helping companies appear in AI chatbot responses—grew from under $1 billion in 2025 to a projected $34 billion by 2034. Early evidence suggests AI referrals convert better than traditional search traffic.

The Question That's Actually Being Asked

The report doesn't land on a thesis. It maps terrain. The gap between capability and deployment is expanding. The gap between leaders and laggards is widening. The economic stakes have moved from speculative to structural. Political volatility is rising, not stabilizing.

What's unresolved is whether the "capability overhang" represents opportunity or risk—and for whom. Organizations slow to deploy face competitive pressure. Those racing ahead face regulatory uncertainty and potential backlash. The tools are production-ready, but the guardrails are still being negotiated in courtrooms and congressional hearings.

The quarter's pattern seems clear: AI isn't waiting for consensus about what it should be allowed to do.

Rachel "Rach" Kovacs is Buzzrag's cybersecurity and privacy correspondent.