Here’s what’s on our plate today:

• 🧠 Cohere story: why ROI-first beats AGI hype.

• 📰 Quick hits on AI listings, Xbox, and runaway agents.

• 🛠 Brain Snack for Builders, specialized AI that actually ships.

• 📊 Poll: Should enterprises chase AGI vision or ROI now?

Let’s dive in. No floaties needed…

The Laboratory

How Cohere represents the alternative to the AGI race

For much of human history, man was obsessed with the idea of flight, and early pioneers dreamt of building a single, universal flying machine that could do everything. However, when the Wright brothers achieved the first powered, sustained, and controlled airplane flight, it sparked a race to build the most powerful and capable aircraft.

As aviation evolved, engineers realized that a single, generalized design would not suffice to meet the demands of different industries. The military required fighter aircraft that were nimble, agile, and capable of executing complex aerial maneuvers. Meanwhile, cargo planes had to be powerful to carry heavy loads, and airliners had to be optimized for comfort and fuel efficiency to make air travel accessible to the general public. Progress, then, came not from generality, but from focus.

Cohere’s alternative AI philosophy

Today, the same tension now defines artificial intelligence. On one hand, artificial general intelligence mirrors the universal-machine dream. At the same time, companies like Cohere reflect aviation’s later lesson that capability and value often emerge from systems designed for defined, real-world constraints.

Founded by three Google Brain alumni (including Aidan Gomez, co-author of the foundational 2017 ‘Attention Is All You Need‘ transformer paper), Cohere represents the counter-narrative to the philosophy that AGI will solve all of humanity’s cognitive problems. According to Cohere, the way to do this is to develop tools that deliver measurable returns for businesses today.

Enterprise AI over AGI

At its core, Cohere builds language models optimized for enterprise retrieval, citation generation, and multilingual capability across more than 100 languages, deployed in environments where data never leaves the customer’s infrastructure. Its North platform, launched in January 2025, allows companies to run AI agents entirely on-premises or in air-gapped environments. The client list reads like a directory of regulated industries: RBC, Dell, LG, Fujitsu, Palantir, and the Canadian federal government.

The company has not received the same traction as its better-known, better-funded rivals, such as OpenAI, Anthropic, and Google DeepMind. It has captured attention through its stated goal, which, according to co-founder Frosst, is to achieve ROI over AGI.

The company’s push for ROI has also attracted investor attention; in August 2025, investors invested $500M in the company, boosting its valuation to nearly $7B. Additionally, the company achieved $240M in annual recurring revenue in 2025, surpassing its $200M target, with quarter-over-quarter growth above 50% and gross margins averaging 70%. While hyperscalers such as OpenAI continue to seek new ways to increase revenue and justify investment, companies such as Cohere are advancing an alternative narrative.

The ROI reality check

For observers of the AI industry, the timing matters. By every external measure, 2025 was supposed to be the year artificial intelligence crossed a threshold. OpenAI had been promising that GPT-5 would represent a generational leap. Sam Altman, the company’s CEO and the industry’s most visible evangelist, had written in January that his team was confident it knew how to build AGI as traditionally understood. The prediction markets agreed: Polymarket gave a 35% chance that OpenAI would formally announce AGI before the year’s end.

AGI expectations meet friction

However, none of that happened; instead, when GPT-5 finally shipped in August, IEEE Spectrum reported widespread user disappointment. Altman himself admitted the company had, in his words, totally screwed up the launch. Days later, in a CNBC interview, he offered a quieter concession: AGI, he said, was not a super useful term. This was a remarkable shift for the man who had built a $500 B company around the promise of AGI.

To make matters worse, Altman appeared to be distancing himself from the very idea that had propelled the industry to invest billions in the pursuit of AGI. Additionally, his statement came at a time when MIT had already published a study finding that only 5% of enterprise AI deployments had reached production, with 95% yielding no measurable return on investment.

Similarly, McKinsey’s annual State of AI survey said that barely 6% of organizations qualified as AI high performers, meaning they could attribute more than 5% of earnings to AI. S&P Global reported that 42% of companies had abandoned most of their AI projects, more than twice the rate from the previous year.

Regulation reshapes competitive advantage

Against this backdrop, Cohere’s positioning looks less like modesty and more like strategy, and a viable one at that.

Another thing strengthening Cohere’s position is the evolution of the regulatory environment. The EU AI Act, GDPR, and emerging frameworks in Canada, the UK, and parts of Asia increasingly favor AI solutions that offer data sovereignty, auditability, and on-premises control. Sovereign AI strategies, in which nations develop domestic capabilities rather than relying entirely on American hyperscalers, are also proliferating in this context. Canada and the UK put it on the right side of this trend.

Beyond the focus on ROI and regulatory compliance, the most telling indicator that Cohere may be on the right path is the talent market’s response to the company. Across the industry, senior researchers are leaving the largest labs, citing exhaustion with the relentless pace, the political dynamics, and the weight of hype. When the people building the technology start quietly migrating toward companies that promise less, that is a signal worth paying attention to.

AI’s shift toward specialization

If the trajectory of aviation has taught humanity anything, it is that flight did not change the world because engineers built one perfect, universal aircraft. It changed the world when the industry accepted that different problems required different machines. Fighters, cargo planes, and commercial airliners succeeded not by trying to do everything, but by doing specific things extremely well.

Artificial intelligence is entering a similar phase. The early years were driven by sweeping promises and the pursuit of ever more general systems. That ambition has not disappeared, but the approach's limits are becoming clearer. Businesses, governments, and regulators are increasingly focused on reliability, cost, security, and measurable outcomes rather than grand visions alone.

In that environment, companies like Cohere represent a different model of progress. While hyperscalers continue to invest in the hope that future breakthroughs will justify today’s spending, Cohere is delivering practical value under real-world constraints. Its strategy reflects a broader shift in the industry from speculation toward application.

This does not mean the AGI race is over or that frontier research has lost its importance. However, AI, like aviation before it, may be moving from its experimental era into a period of maturation. The most lasting impact may come not from one system that can do everything, but from many systems designed with clear, focused purposes.

Quick Bits, No Fluff

• AI listing ‘fails’ inspection: A realtor blamed an AI staging tool after a polished room photo accidentally revealed a filthy real interior in the mirror, spotlighting deceptive listing risks.  

• Xbox gets an AI-era boss: Microsoft replaced Phil Spencer with ex-AI exec Asha Sharma as Xbox CEO while Sarah Bond exits, signaling a contentious reset for the struggling brand.  

• Meta’s agent wrecks an inbox: A Meta AI security researcher says an OpenCLAW test agent went wild in her email, replying to threads and scheduling meetings before she could stop it.

Brain Snack (for Builders)

Wednesday Poll

Meme of the Day

Rate This Edition