The web doesn't need gatekeepers the way you don't need a bank account, driver's license, or a credit card. You can do without it, but it sure makes it harder to interact with modern society. The days of the mainstream internet being a libertarian frontier are more or less over. The capitalist internet is firmly in charge.
The real question is whether there is more business opportunity in supporting "unsigned" agents than signed ones. My hope is that the industry rejects this because there's more money to be made in catering to agents than blocking them. This move is mostly to create a moat for legacy business.
Also, if agents do become the de-facto way of browsing the internet, I'm not a fan of more ways of being tracked for ads and more ways for censorship groups to have leverage.
But the author is making a strawman argument over a "steelman" argument against signed agents. The strongest argument I can see is not that we don't need gatekeepers, but that regulation is anti-business.
You’re totally right, and that’s why I think this era will fail.
Web 2.0 failed because eventually people realized to make money they needed to serve ads, and to do that they needed to own the UI. Making it easy to exfiltrate data meant switching cost was low, too. Can’t have that when you’re trying to squeeze value out of users. Look at the evolution of the twitter API over the 2.0 era. That was entirely motivated by Twitter’s desperate need to make money through ads.
Only way we avoid that future is if we figure out new business models, but I doubt that will happen. Ads are too lucrative and people too resistant to pay for things.
ads really aren’t all that lucrative, though, they’re just simple. I worked for a company that was trying to figure out an alternative to ad revenue (we failed) and our people did some research and the average internet user ends up being shown (if I remember correctly) like $60/month of ads, total.
That’s not surprising and is also so depressing. Imagine what the Internet would be like if there were zero ads and zero of the dark patterns that they essentially force every web site operator to engage in. And all for sixty measly bucks a month that we’d spend in microtransactions or subscriptions or whatever. :(
It’s an average, you can’t just charge everyone in the world and have an ad-free worldwide internet. About 2 billion of internet users are probably worth 0.60$ pm while 200m are worth 600$ of ads pm.
Maybe lucrative isn’t the right word but it’s more profitable than the alternative. Call it simple but there’s only so many ways to monetize people that feel entitled to digital works and refuse to open their wallet.
In practice it'll just mean that each MCP provider will have API tokens and it'll be even harder to lock down spending than AWS. Maybe companies will need to have a company wide system prompt to pretty please don't use the expensive APIs too much.
Obviously it’s possible. And just like the early internet ad networks…
You can’t guarantee they’ll be shown and interpreted correct by the downstream LLM, you can’t guarantee attribution later when a user makes a purchase, you can’t collect (as much) data on users for targeting, etc
The biggest ad networks today (Google, Meta) have strong first party data operations, strong first party attribution techniques, and strong targeting, either through intent (search) or profiles (meta).
MCP originated ads really only threaten Google (via intent based ads), and they’re quickly moving I to owning the UX of LLMs, securing their place in this value chain.
Every time there's an industry disruption there's good money to be made in providing services to incumbents that slow the transition down. You saw it in streaming, and even the internet at large. Cloudflare just happens to be the business filling that role this time.
I don't really mind because history shows this is a temporary thing, but I hope web site maintainers have a plan B to hoping Cloudflare will protect them from AI forever. Whoever has an onramp for people who run websites today to make money from AI will make a lot of money.
There is a difference between blocking abusive behavior and blocking all bots. No one really cared about bot scraping to this degree before AI scraping for training purposes became a concern. This is fearmongering by Cloudflare for website maintainers who haven't figured out how to adapt to the AI era so they'll buy more Cloudflare.
> No one really cared about bot scraping to this degree before AI scraping for training purposes became a concern. This is fearmongering by Cloudflare for website maintainers who haven't figured out how to adapt to the AI era so they'll buy more Cloudflare.
I think this is an overly harsh take. I run a fairly niche website which collates some info which isn't available anywhere else on the internet. As it happens I don't mind companies scraping the content, but I could totally undrestand if someone didn't want a company profiting from their work in that way. No one is under an obligation to provide a free service to AI companies.
To be frank psychiatrists, being MDs, would likely prescribe medication and I’m not sure how that would help. As a licensed psychologist I have ideas on how to debug AI though.
They are literally publishing a book called "If you build this, everybody dies" and trying to stop humanity from doing that. I feel like that's an important detail: they're not the ones trying to create the god, they're the ones worried about someone else doing it.
Managing is a hard job, which is why so many managers are bad and yet, in spite of the efforts of C-levels for decades, the role hasn't been eliminated. There's some wisdom in here, but I would not give this article to a new manager to read. I think it's easy to read too much into this and appear inhuman to your directs.
Is there a line? Sure. Don't shit on your company, but don't do it for your directs...do it for you, because that's just not a healthy way to manage frustration. However, learn to lead in a way that's authentic. Authenticity requires candor.
> Managing is a hard job, which is why so many managers are bad
Many jobs are hard, and I don't think it is a rule that most workers in hard jobs are bad.
My take is multifold:
- Managing takes experience. In the software industry, the evolution has been such that there aren't enough experienced people to fill all the manager positions.
- Startups usually don't hire experience managers because they are deemed too expensive. They end up with inexperienced managers. If it's your first job, you never had a manager yourself, and you're suddenly managing a team, how can you be good?
- It is hard to evaluate the competences of a manager. As an engineer, you can talk with another engineer and get a sense of how good/experienced they are: just ask them to talk about technology. As a manager, it seems harder to evaluate. It's easy to manage a team in a highly functional environment, so you can't say for sure that the manager is good. It's hard to manage a team in a highly disfunctional environment, so you can't say for sure that the manager is bad.
- Managers are promoted from above. It's difficult to judge a manager without considering how their subordinates think about them. I have seen too many people climb the management ladder even though all of their subordinates absolutely hated them.
I've been saying for two years that "any sufficiently advanced agent is indistinguishable from a DSL."
Rather than asking an agent to internalize its algorithm, you should teach it an API and then ask it to design an algorithm which you can run that in user space. There are very few situations where I think it makes sense (for cost or accuracy) for an LLM to internalize its algorithm. It's like asking asking an engineer to step through a function in their head instead of just running it.
I think I understand what you're proposing, but I'm not sure.
So in concrete terms I'm imagining:
1. Create a prompt that gives the complete API specification and some general guidance about what role the agent will have.
2. In that prompt ask it to write a function that can be concisely used by the agent, written to be consumed from the agent and with the agent's perspective. The body of that function translates the agent-oriented function definition to an API call.
3. Now the agent can use these modified versions of the API that expose only what's really important from its perspective.
4. But there's no reason APIs and functions have to map 1:1. You can wrap multiple APIs in one function, or break things up however made most sense.
5. Now the API-consuming agent is just writing library routines for other agents, and creating a custom environment for those agents.
6. This is all really starting to look like a team of programmers building a platform.
7. You could design the whole thing top-down as well, speculating then creating the functions the agents will likely want, and using whatever capabilities you have to implement those functions. The API calls are just an available set of functionality.
And really you could have multiple APIs being used in one function call, and any number of ways to rephrase the raw capabilities as more targeted and specific capabilities.
Evidence that the path to ASI is not extending the capabilities of LLMs, but instead distilling out and compiling self-improving algorithms externally in a symbolic application.
Here are the top articles for the month of May 2023 on HN with "agent" in the title [0]. Looks like early days for the term but with a few large hits (like the HuggingFace announcement), which suggests OP was surprisingly precise in their citation of two years as the time window.
Also, since you're implicitly questioning OP's claim to have been saying this all along, here's a comment from September 2023 where they first said the same quote and said they'd been building agents for 3 months by that point [1]. That's close enough to 2 years in my book.
We need more examples of posts like what you made, where you call it some naysayer for being extremely wrong. Actually, folks can and often do tell the truth on the internet!
I'm sure you can find it in chatbot documentation from the 90s. It's a generic term carried over from non-AI chat. People responding to support chats were called agents.
I also found this take interesting coming from someone at ILM where they grafted Hayden Christensen into Return of the Jedi.
Though in this day and age I can’t help but ask “why not both?” It feels easy to add a choice to your viewing experience. If they can do it for Black Mirror then they can certainly ask up front “which version would you like to see?”
I literally just finished watching Episode IV, the one with the CGI makeover. The extra alien CGI in Mos Eisley is awful. It doesn’t stand up at all, with the one exception of the Jaba scene which gets away with it because it is pretty fun. I wish we’d watched the original version.
Is it easy to find the original? I’d love a copy of each on my Plex server, but I have had trouble finding an original copy. I admit I may not know where or how to look; advice is welcome!
That comparison is really cool. I was mostly paying attention to the 4K77 vs 2011 bluray, and in most cases I thought 4K77 looked better. Not sure why they felt the need to mess with the colors so drastically in the 2011 version.
A 4K fan scan of a 35mm print the was in cold storage since 1980.
It's great to see OG Star Wars looking like in did in '77, with all the optical glitches and the lower contrast with slightly green shadow bias of prints from that time. True time travel that makes the reworked releases look silly.
Another project worth a look is Harmy's fan cuts of the original trilogy, which are tastefully re-assembled from multiple sources and graded.
Nothing the author said is wrong, but I don’t know how much it matters or if it would’ve been better if it handled all this out of the gate. I think if MCP were more complicated no one would’ve adopted it.
Being pretty close to OAuth 1.0 and the group that shaped it I’ve seen how new standards emerge, and I think it’s been so long since new standards mattered that people forgot how they happen.
I was one of the first people to criticize MCP when it launched (my comment on the HN announcement specifically mentioned auth) but I respect the groundswell of support it got, and at the end of the day the standard that matters is the one people follow, even if it isn’t the best.
The real question is whether there is more business opportunity in supporting "unsigned" agents than signed ones. My hope is that the industry rejects this because there's more money to be made in catering to agents than blocking them. This move is mostly to create a moat for legacy business.
Also, if agents do become the de-facto way of browsing the internet, I'm not a fan of more ways of being tracked for ads and more ways for censorship groups to have leverage.
But the author is making a strawman argument over a "steelman" argument against signed agents. The strongest argument I can see is not that we don't need gatekeepers, but that regulation is anti-business.