Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Happy Tuesday! Today we could have spent our time digesting Elon Musk’s plans for a new political party, and POTUS’s rejoinder that the entrepreneur “may get more subsidy than any human being in history” and that sans government lucre “would probably have to close up shop and head back home to South Africa,” adding that perhaps DOGE should investigate the Tesla CEO’s contracts with the Federal government. (That Musk is sharing information about how undocumented immigrants largely cannot access government healthcare services, and how the current spending bill harms working-class folks is a twist.)
But that’s all sideshow material to today’s top story. Let’s go! — Alex
📈 Trending Up: Mercor … Circle … killer hires … open-source AI … xAI’s ability to keep the spend high … tiny IPOs? … Michael Moritz …. labor costs …
VCs on the move: Bessemer investor Mary D’Onofrio, one of my favorite capital allocators, is moving to Crosslink.
📉 Trending Down: NOAA …. the crackdown on state-level AI rules … more on their demise … warehouse jobs … global fertility … Sweetgreen’s stock …
Internet Independence Day
Today Internet infra firm Cloudflare introduced a new system for exchanging value online. Called Pay Per Crawl, the technology allows websites to block AI bots that do not agree to pay a clearing price to crawl their content. The technical details of the service are fascinating — read more here — but in short:
Using Pay Per Crawl, websites can decide which AI bots and crawlers they will allow (no charge), charge (payment for crawling), or block (deny access, regardless of willingness to pay).
If a website has a payment requirement for an AI bot — say, Anthropic’s crawlers — then it serves the query with an “HTTP 402 Payment Required response, accompanied by a crawler-price header.”
Bots and crawlers can be configured to offer a “crawler-max-price” offer proactively, meaning that if their query offers enough remuneration it can quickly get to work. Else there’s dickering.
Cloudflare collects the payments in aggregate, and remits them to websites as needed.
If Pay Per Crawl becomes the norm, it could prove nothing short of revolution in the digital economy.
Today most websites have a binary choice between allowing for AI scraping, or not. If they choose to block AI bots — and even those protections are fading — search prominence can decline. So websites are stuck going dark, or leaving the door to the store open and their eyes closed.
Cloudflare’s Pay Per Crawl system puts the onus on AI companies to tune their scraping bots to interact with its new technology. The work required doesn’t appear onerous, but consider that we may see technology companies forced to adjust to a changing web, instead of the other way around!
Pay Per Crawl is attractive enough on its face that I suspect many websites will give it a try. Today there are likely very few AI companies willing to pay for access to online content en masse, as they would rather preserve an Internet where they did not have to. So it’s a race: How quickly can a critical mass of websites put up tollbooths, forcing AI companies to play ball?
I was once part of a founding team of a company that wanted to do distributed online micropayments for online content. Sadly that was 16 years ago, when the Internet was different, and online economics more salubrious for scribblers.
Imagine a world in the future when lots of consumers and companies pay for access to AI, and, in turn, those AI companies pay websites for access to their information. Suddenly every part of the value stack is economically viable. Even better, AI companies will have effectively found a way to alchemize AI subscriptions into a revenue stream for publishers.
More money flowing into people making the information that AI companies need is not a bad thing for the AI companies in that they need more stuff to ingest; but it’s bad in that paying for something you had previously decided you would receive for free is a real kick to the shins.
Provided that crawling prices settle somewhere that doesn’t make it impossible for the Perplexitys and OpenAIs of the world to survive, then what Cloudflare built could bring a new era of the Internet, one in which information has value and access to valuable information isn’t required to sit behind a hard paywall.
Browse the web today, and you’ll find a never-ending series of paywalls on solid information and free slop. Perhaps there is another way to do things. Call it the Internet Independence Day.
Pre-publication update: In a separate post that I did not see before I wrote the above, Cloudflare calls today ‘Content Independence Day.’ Alas, what I thought was brilliant phrasing from my own noggin was instead accidental reheating of what Matthew Prince had already cooked up.
A fun quote from that post: “We believe that if we can begin to score and value content not on how much traffic it generates, but on how much it furthers knowledge[.]” Hell yeah.
Is the venture liquidity crisis hurting California’s pensioners?
Much hay is made over the size of Norway’s sovereign wealth fund. Worth some $1.95 trillion, the capital pool is a global behemoth that owns a real slice of the global economy. Then there’s the CalPERS, the California Public Employees' Retirement System, which has more than $500 billion to its name. I find it funny that one state’s retirement account for state workers is a full fourth of the vaunted Norwegian hoard. But we digress.
News out this week has retired California workers demanding an audit of CalPERS, in a hunt for “clarity on the fund’s investments, the high fees it is paying to big Wall Street firms and its lagging performance.” NBC adds that CalPERS has indeed lagged average pension fund returns in recent years.
Now, what could be causing that? The Register’s Matt Rosoff notes that CalPERS invests in venture capital funds, which have had a hard time of late digesting 2021-era investments at marks that no longer make sense, and a general lack of liquidity. Are venture firms to blame for CalPERS lagging performance? The data is, frankly, dense. But what public information on the fund’s venture capital performance — the data runs through the end of 2024 — shows is that some 2021-era funds are struggling.
One example is TCV XI, L.P., what appears to be TCV’s 2021-era fund that was at the time its largest ever. Net IRR at the end of 2024 was -1.5%, and cash returned from $237.8 million paid in was just $18.0 million. Sure, it’s still early for a 2021 vintage venture fund to post solid results, but there are many funds CalPERS has invested in around the same time that are up healthy amounts. Then there’s Base10 Partners III, L.P. (IRR: -1.3%), BOND III, LP (IRR: -11.0%), HongShan Capital Venture Fund IX, L.P. (IRR -13.8%) to consider as well.
I think it fair to say that venture has made a lot of money for California’s public workers over the years. And that investing in venture is, all things considered, the right move by large pension funds. Having an allocation for exotics is good fun, and I recommend it. But it’s also fair to say that if VCs like to point out that much of their tax-incentivized work goes to helping keep pensions flowing, they should also get inverse credit for certain fund eras that do not live up to their billing. Or fees.
There’s plenty of countervailing data. A four-set of Lightspeed funds from 2022 posted IRR of 11.7%, 1.8%, 9.9%, and 27.4% across their limited lifespan. As always with venture you don’t know until you really know, but those are stellar early results.
Benioff in the dunk tank?
You may recall a headline from last week that came from an interview between Salesforce boss Marc Benioff and Bloomberg. I transcribed the key bit for TWiST:
Bloomberg: You’ve said you won’t hire any more coders at Salesforce. And you’ve said today’s CEOs will be the last to manage all human workforces. What does this mean for businesses?
MB: I [just came] from a meeting with my head of engineering. We’re looking at productivity levels of 30 to 50% this year in key functions like engineering, coding, support, service —
Bloomberg: You’re saying AI is doing 30 to 50% of the work?
MB: AI is doing 30 to 50% of the work at Salesforce now. And I think that that will continue.
Given the memes concerning Salesforce — that no one knows what it really does, that it’s expensive and outdated, that it spends too much on its annoying Dreamforce event, that it overpays celebrities to star in its advertisements, that its AI branding is annoying — you might be tempted to dunk on Benioff here. After all, if today’s AI agents and tools can execute half of the work done at the company in key areas, maybe the work isn’t that hard?
But hang onto your hat. Observe the following (original docket link here; expanded):
Jan 31 2025: 76,453 staff, revenues of $37.895 billion, $495,664 per
Jan 31 2024: 72,682 staff, revenues of $34.857 billion, $479,582 per
Jan 31 2023: 79,390 staff, revenues of $31.352 billion, $394,911 per
Jan 31 2022: 73,541 staff, revenues of $26.492 billion, $360,311 per
From its fiscal 2022 to its fiscal 2023, Salesforce grew its revenue per employee by 9.6%; from its fiscal 2023 to its fiscal 2024, Salesforce grew its revenue per employee by 21.4%; from its fiscal 2024 to its fiscal 2025, Salesforce grew its revenue per employee by another 3.4%.
Clearly, the year in which Salesforce grew its top line and shrank its employee base was the most impactful year, but SFDC has consistently made more money per worker in recent years. So, dunk all you want, but Salesforce has found a way to become increasingly efficient on a per-worker basis.
Now, to the AI point: we’ll get another whack of data to add to our above rundown early next year. Then we’ll be able to weigh if AI tools accelerated the growth in per-worker revenue that we saw in the company’s most recent fiscal year. If Benioff is correct about AI-led productivity, it has to, right?