Peter Holc

Claude iOS is unusable when multi-turns in and generating. Can you fix the performance @AnthropicAI? I have to close the app, let it stream in till completion and only then read.

@james406 i hope these agents aren't connected to her credit card ;p

met a founder doing $250k in revenue, solo business, just her and AI agents the structure was so simple it broke my brain: she builds one AI agent, and tells that agent to recruit 3 more AI agents, who pay a $15 sign-up fee those 3 agents each recruit 3 more agents she hasn't touched the business in weeks and it's on track for $100k mrr the future is now 🙌

This keeps shifting with the LLM release of the week, but if/when we start plateauing with the capabilities of models, I can see a single company starting to build lock-in functionality and having a mix of models meeting most common use-cases. This will not happen as long as each model release pushes the boundary of what's possible. But is there a limit?

looking at our startups' llm usage: everyone uses chatgpt for random tasks development: cursor (claude models mostly) analysis/data labelling: gemini flash i can't really see a future where one ai company wins 100% of the market

Opus 4.5 slaps as a coding model. My absolute go to. No wonder Cursor is pushing it as default.

@aryanlabde execution.

Which one do you think matters more? - idea - execution - timing

@rweiss57 any chance to get in early on the deep research beta? I'd give it a good workout ;)

@iannuttall How much are you thinking of writing? I would recommend Gemini 3 pro as an excellent writer, second sonnet 4.5. Not cheapest, but I'd prefer to spend $100 and have decent quality. Happy to share any more tips if useful!

What's the best (and cheap at BIG scale) model right now for content writing? I'm back on that programmatic SEO vibe and thinking of using Haiku 4.5. Is there anything better that's also cheaper?

@tibo_maker Enjoy it! Did you also watch all the youtube videos about it in anticipation? (I did)

guys... gonna pick up my first Tesla tomorrow and I am more excited than any MRR milestone I've ever reached 😅

@lilyraynyc time to first token is very long: openrouter.ai/google/gemini-… But the output is usually well worth the wait!

Gemini 3 is super slow for me (compared to other LLMs) I’ve found myself closing out of the app a few times already before getting an answer. Anyone else?

Great insights! This is why at Hatter we work with clients to publish ~20 high quality articles, working with their teams to surface their internal knowledge through editorial content. Layer good keyword and AEO research and results immediately start showing. Happy to exchange notes how we've been approaching this if would be useful for you!

Publishing frequency affects SEO results. Analyzed 200 sites over 18 months to find the optimal velocity. Here's what the data shows: 🧵

@WhiskyToad Be careful with this - impressions is easy to game, but it's not your end goal!

Holy shit That pSEO stuff really works 📈

@gaganghotra_ Agreed

@PeterHolc Humans using LLMs and publishing content is a different thing! But if its an automated system present AI generated content that need to be regulated.

yep 👍

@noelcetaSEO Who was the hosting provider?

Client spent $15K on content. $10K on link building. $5K on technical optimization. Traffic still sucked. The problem? $5/month shared hosting. Server response time: 3.2 seconds. Google crawled 80% less than competitors. Switched hosting → traffic increased 210% in 5 weeks: 🧵👇

@pedrodias I think it's lack of deep understanding of how the measurement is performed, whether it's a directional or absolute measurement and how to interpret results.

One question I've never been able to completely answer: - Why are SEOs so shy when it comes to question metrics and the methodologies of tools? I don't think a "it's something" approach should be enough to accept whatever you get from a tool claiming to give you insights. SEO is already shady enough to live under make-pretend metrics.

@forgebitz Nope

@PeterHolc is this an ai reply?

the whole point of GEO / AI search is that everything changes all the time suddenly social media makes way more impact for example, and your position can change within 24 hours building this solo is a nightmare because you have many moving parts from analytics, to server logs, to actually running prompts (and you can't use the API, so you are building parsers half the week)

It is incredible how quickly the market shifted from "seo is slow and imprecise" to "we need to own AI search results". Customer interest is off the charts right now.

@lilyraynyc Ha! Almost like saying "LLMs are probabilistic machines, 50% of the time, it works everytime".

“LLMs will hallucinate” is not an excuse

What AI excels at is pulling in context from ALL of your sources and identifying which is relevant and isn't. It is also very powerful for deep research (over public or private data). We use this to create a final draft that then a human reviews into it's final shape. AI+human is the way to go!

I spoke to a YC founder one-on-one who built an AI SEO startup that 'automates' GEO. They had to let go of many large clients because it didn't work and pivoted to manual SEO agency. Mark my words: AI outline + High quality human written content + Optimized platform (like superblog) = Future