Boyuan :D

Make Websites Grow Themselves @ RankAI (YC Backed) | Unlock millions of buyer from Google & ChatGPT | Forward Deployed Founder

[Boyuan :D @boyuandeng]

@shengkun_ye Kinda cool. Wanna try u guys out!

[Boyuan :D @boyuandeng]

@hanghuang_ What’s an awesome journey fellas. Heard a lot of great things from talking to many investor. Congrats on the raise and the good stuff is yet to begin!

[Boyuan :D @boyuandeng]

We run loops like this every day for SEO delivery. A trace catches the redundant tool calls and the errors just fine. It won't catch the run that came out technically correct and completely useless to the customer, because usefulness isn't a field in the trace.

[Guillermo Rauch @rauchg]

Agentic self-improvement.

Give your agent the ability to introspect its past runs, spot inefficiencies, errors, redundant tool calls, and produce new prompts and skills.

That’s why agent observability is built-in when you deploy https://t.co/99eEa13mZ3 to Vercel.

[Boyuan :D @boyuandeng]

Cloudflare published the ratio: ClaudeBot crawls 23,951 pages for every one visitor it sends back. Bots just passed humans: 57.5% of HTML requests.
The biggest reader of your site no longer clicks.

[Boyuan :D @boyuandeng]

Seer tracked 53 brands across 5.47M queries. Brands cited in AI Overviews saw organic clicks lift 35% and paid lift 91%.
Being the answer is starting to beat being the first result. That has been the whole bet at RankAI, and it lands differently when the receipts come from someone else's dataset.

[Boyuan :D @boyuandeng]

@rajshamani Agree. Though the judgment I've watched actually decide things shows up late, weeks in, when the demo looked great and the customer stops being impressed by motion. How do you spot that kind of judgment in someone before their work has had time to age?

[Boyuan :D @boyuandeng]

@esthercrawford The time is what does it. You spend real hours with someone good, you start rooting for them, and by the no it stops feeling like a hiring call and starts feeling like you let a person down. Weirdly, I keep thinking about the strong ones long after the no.

[Boyuan :D @boyuandeng]

@emollick We run this in our SEO delivery loops. The delegating works, the trouble is the planner trusting a cheap model's output without re-checking it. Have you seen the planner learn when to distrust its own delegate, or do you still gate the cheap steps by hand?

[Boyuan :D @boyuandeng]

Agents can make a rep 5x faster at sending touches. That was never the part that closed for us. The only channel that consistently worked was getting on a plane and pitching ICP in person, and I don't see agents shrinking that.

[Jason ✨👾SaaStr.Ai✨ Lemkin @jasonlk]

"What will sales teams look like going forward?

Probably half the size they used to be, pre-AI.

And reps might even be 5x as efficient in a few years." with @samdblond

[Boyuan :D @boyuandeng]

Same story in the SEO tool space. Everyone's got the pretty dashboard and the line going up on the sales call. The tools still alive a year later are the ugly ones grinding the delivery nobody screenshots.

[swyx @swyx]

i often think about the irony of how "tools for thought" people spent like a decade making cool pretty demos with canvases

and then got completely mogged by low contrast poorly designed CLIs just winning because they do commodity thinking for you

[Boyuan :D @boyuandeng]

Routing rules solve the sudden retirement problem, and that part holds up. What I haven't seen solved anywhere is output consistency. Same route, different model, and the pages your agents produce start reading differently, no gateway rule catches that.

[Guillermo Rauch @rauchg]

We've been explaining AI Gateway as a C̶o̶n̶t̶e̶n̶t̶ Token Delivery Network. Like a CDN for AI models.

One great feature of CDNs is the ability to dynamically re-route or deny traffic without redeployment.

When Fable was suddenly retired, we worried about production workloads depending on it. Even Fable aside, models get retired quite often as GPU capacity is heavily contested. Yet our data shows people still having happy production traffic of older model versions!

We're solving this now with AI Gateway Rules. e.g.: you can now rewrite model "routes" on the fly!

𝚟𝚎𝚛𝚌𝚎𝚕 𝚊𝚒-𝚐𝚊𝚝𝚎𝚠𝚊𝚢 𝚛𝚞𝚕𝚎𝚜 𝚊𝚍𝚍 \
--𝚝𝚢𝚙𝚎 𝚛𝚎𝚠𝚛𝚒𝚝𝚎 \
--𝚜𝚘𝚞𝚛𝚌𝚎 ���𝚗𝚝𝚑𝚛𝚘𝚙𝚒𝚌/𝚌𝚕𝚊𝚞𝚍𝚎-𝚏𝚊𝚋𝚕𝚎-𝟻 \
--𝚍𝚎𝚜𝚝𝚒𝚗𝚊𝚝𝚒𝚘𝚗 𝚊𝚗𝚝𝚑𝚛𝚘𝚙𝚒𝚌/𝚌𝚕𝚊𝚞𝚍𝚎-𝚘𝚙𝚞𝚜-𝟻

This adds to the repertoire of capabilities and techniques AI Gateway brings to the table to recover lost tokens. If you drop tokens, you lose revenue and customers 💸

Kudos to @rtaneja_ @crowprose @shaper – we went from CLI design to production-grade capability very quickly!

[Boyuan :D @boyuandeng]

Watched a small version of this from inside GEO. llms.txt went from must-have to does-nothing inside a year, and the actual Google index barely changed the whole time.

[Marc Andreessen 🇺🇸 @pmarca]

The great American techno-capital dynamo goes brrrr. 🇺🇸

[Boyuan :D @boyuandeng]

In 2020 every broadband company had a guy who drove over to set up your router. Decent pay, needed everywhere. Then setup became an app and the house calls ended.
FDE reads the same to me. The job is carrying context into companies that can't hold it yet, and it pays until they can. Still the best job for watching where agents actually break.
I guessed 2-3 years back in May. Might be shorter.

[Boyuan :D @boyuandeng]

A 67 story tower with 1,019 apartments, approved, in San Francisco. The city that was famous for saying no. Time to build!!!

[Garry Tan @garrytan]

It’s time to build

[Boyuan :D @boyuandeng]

A page can rank #1 now while the AI Overview answers the question right above it.
The click never happens. The ranking report still says you won, the visits just never show up.
Rank was always a stand-in for getting seen, and AI answers split those two apart.

[Boyuan :D @boyuandeng]

@Ric_RTP $28B monthly volume shows prediction markets have matured into serious financial instruments. If Kalshi stays independent, it could drive more innovation in event contracts, but competing against Meta's resources will test whether regulatory moats truly protect smaller players.

[Boyuan :D @boyuandeng]

@jasonlk Running agent loops daily for SEO work and the waiting trap is real. Kick off a run, drift, come back having done none of the thinking the agent can't do for you. Happens inside a single workday, never mind between people.

[Boyuan :D @boyuandeng]

@emollick GEO sits right on that line for us. Whether a page ranks you can verify, whether it deserved to rank is a judgment call.

[Boyuan :D @boyuandeng]

There was a guy in the 2000s whose whole job was coming to your place to set up the internet.

Real skill, real demand.

Then setup became self-serve and the visits quietly stopped. That's how I see the FDE wave.

The gap between AI-native teams and traditional companies is real, which is why it might be the best seat in tech right now.

But the job itself is closing the gap that pays for it. I'd still take the seat. Just wouldn't plan to sit in it for ten years.

[Boyuan :D @boyuandeng]

@snowmaker True evolution. Just like responsive design unified web/mobile, we'll need "agent-native" architectures—declarative goals + observability layers that serve both humans and LLMs seamlessly.