Agentic AI, meanwhile, arrives at the intersection of profit, technology, and politics

By Jennifer Evans, Pattern Pulse AI / B2BNN / Tech Reset Canada

Every day brings a new framework, a new benchmark, a new “10x” launch. The issue stops being how to keep up and becomes: which signals are real and which are noise, wearing the costume of urgency. A post by Rohit on Twitter made the rounds this week with a clean version of a response: the field has no destination, the giants don’t know either, the people winning in two years picked durable primitives early and let the rest pass them by. And all the while, the acceleration continues. I won’t stop because it isn’t linear or centralized. The acceleration is everything everywhere, all the time. As both the people inside and watching on the periphery, it’s impossible to say where the next important signal will come from, only that several noteworthy developments will pass you by completely before you’ve digested the most recent.

That part is right. What it leaves out is harder to see. Most of the loudest signals in AI right now are pointing one layer off from where the value is actually being captured, and one layer off from where the political fight has already started. Two examples are running this month, and both illustrate the same failure mode.

The CPU misread

A widely shared post this week, citing Andrej Karpathy’s Software 3.0 talk, declares the CPU finished as the main chip. The framing: the FPU got swallowed by the CPU in 1989, the CPU got demoted by the GPU after CUDA in 2007, and now the third flip is happening, with the model itself replacing the chip as the seat of value. Intel sits at $425 billion. Nvidia hit five trillion. The gap is the story.

The data refutes the framing. Intel posted Q1 2026 revenue of $13.6 billion, beating its own guidance midpoint by $1.4 billion, with the Data Center and AI group up 22 percent year over year. Intel hit a 25-year stock high. AMD is at an all-time high. Intel said on the same call that the CPU-to-GPU deployment ratio is tightening from 1:8 to 1:4 and could reach 1:1 in agentic scenarios. Frontier labs are running out of CPUs for RL training and competing directly with cloud providers for x86 allocation. Microsoft is selling spare CPU inventory to Anthropic and OpenAI because the model labs need so many CPUs that Microsoft is rationing.

That last data point keeps getting read backwards. People read it as evidence the CPU is dying. It is evidence the CPU got pulled back to the centre of the agent economy. Every tool call, every RAG hop, every orchestration step lands on a CPU. Agents are CPU-hungry by architecture.

Karpathy’s actual framing was metaphorical. The LLM is the CPU of a new computing paradigm, with the context window as RAM and prompts as programs. The X version collapsed that metaphor into hardware displacement and missed the layer the metaphor was operating on. The value did move up the stack. It moved into the model. The substrate underneath got amplified.

The agent misread

“The year of agents” has been declared three years running. The current headlines: Claude Code authoring 4 percent of public GitHub commits with a projected 20 percent by year-end, Cowork shipped by Anthropic in ten days, OpenClaw integrating DeepSeek V4 with full local system access. Real numbers in coding. Real adoption.
What the agent discourse will not look at is the only other agentic AI application currently deployed at industrial scale: surveillance pricing.
Forty-plus state bills across 24 U.S. states have been introduced in 2026 already, outpacing all of 2025. The House Oversight Committee launched a formal investigation on March 5. The FTC testified in April that staff work on surveillance pricing continues.

Consumer Reports ran a field test across 400 volunteers and found that 74 percent of Instacart items were offered at multiple price points simultaneously, with some shoppers seeing prices 23 percent higher than others for the same item at the same store. The estimated invisible tax on a family of four was $1,200 a year. The FTC’s January 2025 staff findings confirmed the explicit pitch behind the industry: vendors market these tools as a way to boost revenue and margins by two to five percent through extraction.
That is the production agentic economy hiding behind the demo theatre. Pricing engines that watch consumer behaviour, infer willingness to pay, and adjust the number on the screen accordingly. Mouse movements. Demographic inference. Abandoned cart history. Browser fingerprints. The infrastructure exists, the deployments are live, and the revenue is real. The Claude Code adoption curve is real too. The centre of gravity in deployed agent revenue outside of coding is consumer extraction.

The Canadian fight is already underway

The most striking thing about Canadian AI discourse right now is that the political fight on surveillance pricing has already started, and most of the AI policy world is looking the wrong direction.
Avi Lewis won the federal NDP leadership on March 29 on the first ballot with 56 percent of the vote, the largest margin in NDP history. Two weeks later, on April 13, he stood with his caucus in Ottawa and announced a motion in Parliament to ban surveillance pricing, calling it “a crystal clear example of why we desperately need government guardrails to protect us from the triple threat of Big Tech, AI and corporate monopolies.” The framing he chose for his first major policy push as leader was the AI consumer extraction layer.

The lineage matters. Manitoba’s NDP government became the first jurisdiction in Canada to introduce a surveillance pricing bill. The Ontario NDP under Marit Stiles followed. UFCW Canada is publicly backing the federal motion through its national president. The grocery and retail labour movement is in the same room as the AI policy push, which is unusual and significant.

The refusals matter more. Doug Ford, asked directly whether he would ban surveillance pricing in Ontario, said he would not because he believes in free-market capitalism. The Carney Liberal majority government has not moved on the file. The two governments with the actual authority to act, federally and in the largest province, are holding the line for the practice while the only deployed-at-scale extraction infrastructure outside of coding agents continues to run on Canadian consumers.

That is a sovereignty story sitting in plain sight. Federal AI policy attention is consumed by the Sovereign Compute Infrastructure Program and the question of where GPUs physically sit. Provincial attention, in Ontario at least, is consumed by the medical records announcement of March 2026 with no funding and no timeline. The deployed agentic AI economy is already running on Canadian grocery carts and travel platforms and rental listings, and the political response is split clean along party lines, with the NDP advancing, the Liberals silent, and the Ontario PCs defending the practice on principle.

What the misreads have in common

Both are layer errors. With CPUs, the discourse stares at hardware while value capture moved into the model. With agents, the discourse stares at coding agents and demo theatre while the production deployment is consumer extraction. The pattern: the loudest signal sits consistently one layer off from where the action actually is, and the political fight starts in the place the discourse is not looking.

The filter

Rohit’s tests hold up. Will this matter in two years. Has someone you respect lost a weekend to it. Does adopting it force a migration. What does skipping for six months actually cost.

Add one more for the agent era. Where is the money actually being made right now, who is paying it, and which government is defending the practice. The answer today is uncomfortable. Coding agents make money from developers. Surveillance pricing engines make money by extracting from consumers who do not know the price they see was built for them. In Canada, the federal Liberals and the Ontario PCs are the ones declining to intervene.

The professional skill of this field is the willingness to look one layer deeper before reacting to the launch. Skip the demos. Read the postmortems. Watch the state bills, the provincial bills, the parliamentary motions, the procurement filings. The shape of the deployment, and the shape of who is willing to defend it, is visible there before it shows up in the keynote.