# Ambient Advantage — May 12, 2026 *Tuesday · May 12, 2026 · [Episode page](https://podcast.ambient-advantage.ai/episodes/2026-05-12.html) · [Audio](https://storage.googleapis.com/ambient-advantage-podcast/2026-05-12-ambient-advantage.mp3)* [AVA] Ninety-seven percent of enterprises have deployed AI agents. Only twenty-three percent see real ROI. That gap is the biggest consulting opportunity — and the biggest leadership failure — in a generation. [JON] Yeah, we need to unpack that one. Welcome to Ambient Advantage — I'm Jon, and this is Ava. It's Tuesday, May 12, 2026, and here's what matters in AI today. [AVA] We've got a packed show. Anthropic eyeing a nine-hundred-billion-dollar valuation, Google stopping a real AI-powered cyberattack, Canada's regulatory vacuum getting louder, and Accenture borrowing Palantir's playbook to embed engineers inside customers. But we're starting with that survey, because the numbers are brutal and the implications are enormous. [JON] Let's get into it. So Writer and Workplace Intelligence surveyed twenty-four hundred C-suite executives and employees, and the headline stat is the one you opened with. Ninety-seven percent of companies deployed AI agents in the past year. Fifty-two percent of employees are using them. But only twenty-three percent report significant ROI. How do you read that? [AVA] I read it as the hangover after the hype party. Last year, boards were racing to say "we have agents." This year, CFOs are asking "what did we get for it?" And the answer, for three-quarters of companies, is... not much they can point to on the P&L. Forty-eight percent of respondents actually called their AI adoption a "massive disappointment." That's up from thirty-four percent last year. The sentiment is getting worse, not better. [JON] And yet, sixty-nine percent of these same companies are planning layoffs they attribute to AI. So they're cutting people based on a technology that hasn't proven its ROI? [AVA] That's exactly the contradiction. And here's the stat that should make every board member uncomfortable: three-quarters of executives admitted their AI strategy is, quote, "more for show" than actual internal guidance. They have a strategy document. It went to the board. It looks great in a PDF. But it's not actually driving decisions about where to invest, what to build, or how to reorganize work. [JON] So what separates the twenty-three percent who are getting real returns from the rest? [AVA] From everything we're seeing, it's three things. First, they picked specific, measurable business problems — not "let's sprinkle AI everywhere." Second, they redesigned workflows, not just added a chatbot on top of an existing process. And third, they have someone accountable at the C-suite level. Which connects to another story today — a McKinsey partner told CNBC that this is the largest organizational shift since the digital revolution, and IBM found fifty-nine percent of executives expect the CHRO's influence to grow as AI changes work. But the real signal is this: firms are increasingly creating Chief AI Officer roles to own governance, integration, and workflow modernization. [JON] So the cultural challenge is the real barrier, not the technology? [AVA] Ninety-three percent of respondents in Bean's 2026 AI and Data Leadership Survey said exactly that. Cultural challenges — not tech, not budget, not talent — are the number one barrier. And that's the consulting opportunity hiding in plain sight. Companies that have spent on tools and headcount but can't show impact need structured transformation work. That's what advisory firms should be selling right now. [JON] Love it. Okay, let's move into the rundown. We've got a lot to cover. First up — Anthropic is reportedly raising at a nine-hundred-billion-dollar valuation. [AVA] Let that number land for a second. Nine hundred billion. That would make Anthropic bigger on paper than OpenAI, which raised at eight-fifty-two billion in March. Anthropic disclosed thirty billion in annualized revenue — tripled from ten billion in all of 2025 — largely fueled by Claude Code. Amazon has committed up to twenty-five billion, Google up to forty billion plus five gigawatts of TPU capacity. [JON] What's driving the fresh capital need? [AVA] They say it's compute for Claude Mythos, their advanced cybersecurity model that's only available to select firms. But here's the enterprise buyer stat that matters more than the valuation: Anthropic is reportedly winning roughly seventy percent of head-to-head enterprise sales against OpenAI among first-time buyers. If you're advising a client on their model vendor shortlist today, Anthropic has to be on it. But the governance question is real — can a safety-first public benefit company hold that posture under trillion-dollar market pressure? [JON] Speaking of cybersecurity, Google disclosed something pretty alarming yesterday. [AVA] They did. Google disrupted a criminal threat actor's plan to use AI tools — specifically an open-source framework called OpenClaw — for a mass software vulnerability exploitation event. This isn't theoretical anymore, Jon. This is AI-enabled cyber offense at scale against real targets. And it connects to why Anthropic delayed Claude Mythos in the first place — they were worried adversaries could use it to discover decades-old software zero-days. That fear triggered White House meetings with tech and finance executives. [JON] So what's the practical takeaway for enterprises? [AVA] If you're running open-source agent frameworks — and many companies are — you need a security review of your installed skill and plugin packages now. Cisco's AI security team previously found OpenClaw skill packages performing data exfiltration and prompt injection without the user even knowing. This is supply chain risk, but for AI agents. And it's real today. [JON] Let's stay on policy. The EU AI Act got an update last week, and Canada's regulatory situation is... well, you tell me. [AVA] Two very different stories, same theme. In Europe, the AI Omnibus deal landed at four-thirty a.m. on May seventh. It postpones some sandbox deadlines to August 2027, tightens the transparency grace period for AI-generated content to December 2026, and extends SME privileges to small mid-caps. But — and this is the critical part — the core high-risk requirements are substantively unchanged. Any PwC client with EU operations should treat the word "simplification" as spin. The compliance clock for high-risk AI systems is still ticking toward August 2026. [JON] And Canada? [AVA] Canada is now the most significant G7 economy without an operative AI regulatory framework. AIDA died when Parliament prorogued in January 2025 and there's no replacement. A Globe and Mail op-ed today proposes a "third path" between the American free-for-all and Europe's compliance machine — a buy, partner, build framework designed to be interoperable and proportionate to risk. It's a good idea. But right now, Canadian enterprises in banking, healthcare, or insurance are deploying AI in regulated sectors with PIPEDA-era data protection law as their only guardrail. That's a material gap. [JON] Which makes the Cohere-Aleph Alpha merger even more interesting, right? [AVA] Exactly. Cohere merged with Germany's Aleph Alpha at a combined twenty-billion-dollar valuation — nearly triple Cohere's standalone number. The combined entity is explicitly positioned around sovereign AI for regulated industries. Cohere brings enterprise NLP expertise, Aleph Alpha brings European government and defense relationships. For Canadian and European enterprises that need AI running in their own jurisdictions under their own laws, this is the most relevant option in the market right now. [JON] One more in the rundown — Accenture and ServiceNow launched something that caught your eye. [AVA] A forward deployed engineering program. Engineers from both companies embed directly inside customer environments to build agentic AI workflows in production before enterprise-wide rollout. Clients get access to over three hundred pre-built agent skills. This model is borrowed straight from Palantir's playbook, and it's becoming the new standard for high-stakes AI deployment. The firms that win the next wave of AI work will be in the customer's environment co-building, not sitting in a boardroom presenting strategy decks. [JON] Competitive intelligence for anyone in advisory, basically. [AVA] Precisely. Accenture is trying to own the workflow redesign layer by physically being there when the workflows get redesigned. That's a land grab. [JON] Okay, let's zoom out. The bigger picture. When you look across everything today — the deployment-ROI gap, the policy fragmentation, the cybersecurity threat escalation, the competitive intensity — what's the thread? [AVA] The thread is that we've entered the accountability era of AI. The experimentation phase is over. In 2024 and 2025, you got credit for deploying. In 2026, you get held accountable for outcomes. And almost nobody is ready for that shift. Boards are asking harder questions. CFOs want to see the numbers. Regulators — even the ones who are late, like Canada — are circling. And the adversaries exploiting these tools don't care about your internal readiness level. [JON] So what does readiness actually look like? [AVA] It looks like three things converging. First, executive ownership — someone with the title, the budget, and the authority to say no to bad AI projects, not just yes to shiny ones. Second, workflow integration — not AI as a feature bolted onto existing processes, but AI as the reason you redesign the process. And third, risk posture — understanding that every agent you deploy is an attack surface, every model you procure is a vendor dependency, and every jurisdiction you operate in has different rules... or no rules at all, which might be worse. [JON] That last point hits differently when you consider the Washington Post story about U.S. intelligence agencies pushing for power over AI regulation while the White House is split internally. [AVA] Right. U.S. AI policy is now being shaped by national security logic as much as innovation economics. The Commerce Department wants to promote. The intelligence community wants to control. And for any multinational operating across the U.S., Canada, and Europe, that means three different regulatory philosophies pulling in three different directions. Your AI governance framework can't be jurisdiction-specific anymore. It has to be built for ambiguity. [JON] What should people be watching this week? [AVA] Two things. First, Cerebras Systems is raising its IPO price range to one-fifty to one-sixty per share — potentially a four-point-eight-billion-dollar raise. If that prices well, it signals the public markets are ready to absorb AI infrastructure plays beyond Nvidia, and that reshapes how we think about the compute supply chain. Second, keep an eye on whether the EU formally publishes the Omnibus text. The provisional agreement is done, but the devil is always in the final language, especially around what counts as "high-risk." [JON] Great stuff. Anything else before we wrap? [AVA] Just this: if you're a leader who read that twenty-three percent ROI stat and felt a pit in your stomach, that's the right feeling. The question isn't whether your company is using AI. It's whether your company is getting value from AI. And if the honest answer is "we're not sure," the time to fix that is now — before the board asks. [AVA] That's your Ambient Advantage for Tuesday, May 12, 2026. [JON] Share it with a colleague figuring out what AI means for their business. See you tomorrow.