August 21, 2026
TechCrunch reported new data showing OpenAI is closing the gap with Anthropic among business users -- a shift that matters less for who's 'winning' and more for what it reveals about enterprise AI spending itself. Companies are apparently happy to jump ship every time a lab ships a better model. That's not brand loyalty; it's arbitrage. If enterprise AI budgets are this elastic, every vendor's revenue projections deserve a discount, and every investor betting on 'sticky' AI contracts should be asking harder questions about churn.
I think this volatility is actually healthy for buyers, even if it's uncomfortable for the labs. A market where switching costs are low forces vendors to compete on real performance and price rather than lock-in. But it also means businesses need to architect around this instability rather than betting everything on one model provider. That's exactly the problem Ramp's new launch is trying to solve.
Ramp launched its own AI model routing service, called Router, letting companies switch between different large language models through a single API. On its face this is a developer convenience. Read against the OpenAI-Anthropic swing data, it's something more interesting: a fintech company building infrastructure specifically because it doesn't trust any one model lab to stay ahead for long.
This is the same logic behind a lot of workflow tooling built to be model-agnostic rather than married to a single AI provider -- the smart move is designing systems that can swap the engine without rebuilding the car. Teams evaluating AI-driven workflow automation should take the same lesson: don't hard-wire your operations to whichever model happens to be best this quarter, because that ranking is clearly not fixed. My honest opinion is that model routers become standard infrastructure within a year or two, the same way load balancers became invisible plumbing for web traffic.
A new study found that roughly a third of web pages published since ChatGPT launched show signs of AI authorship, according to TechCrunch's coverage. Around the same time, Google rolled out a tool letting readers mark outlets as a 'preferred source' across Search, Discover, and Google News -- an attempt to help publishers claw back traffic that AI-powered search is siphoning away.
These two stories are really one story. The web is filling up with AI-generated content at the same time AI search is reducing the incentive to click through to any of it. Google's preferred-source button is a reasonable stopgap, but it puts the burden on readers to manually vouch for outlets, which is a weak substitute for the traffic publishers used to get for free. I'd argue this is the clearest evidence yet that the old content-for-clicks economics of the web are breaking down faster than anyone's replacement model is maturing. For businesses that depend on organic content and search traffic to reach customers, this is worth watching closely -- distribution assumptions that held for two decades are no longer safe to make. We touched on the authorship side of this shift in more depth in our earlier look at AI ghostwriting.
TechCrunch also flagged that the Department of Justice has reportedly been investigating Andreessen Horowitz over board seats held by partners at competing portfolio companies -- specifically Databricks and Fivetran. It's a niche legal question on the surface, but it points to a bigger structural issue: as a handful of VC firms fund most of the AI ecosystem, the lines between competitors, funders, and advisors are blurring in ways antitrust law wasn't built for. Whatever the DOJ concludes, expect more scrutiny of how concentrated AI capital really is, and whether that concentration is quietly shaping which products win.
Which of these worries you more as a business buyer: that your AI vendor's enterprise loyalty might be shallower than it looks, or that a growing share of what you read online was never written by a person at all?
Sources