It's not available in where I live (Prague).

Also not as comprehensive as I'd need.

Searching for a flat to buy...

Whenever there's a new listing, it goes through my filters which aren't easy (orientation based on flat layout in the photos, light based on photos, distance to my son's school using Google directions API), and notifies me.

Saves me easily 30-60 min a day.

Agents are Commoditized Here's what I think is not commoditized

I finally made a free substack and wrote my first article there. I'm building agents every day, so wanted to share some learnings there.

alpergondiken.substack.com/p/agents-are...

Yes but it'd be a small profit (relatively) and investors who poured billions wouldn't be happy with the valuation of the company whose profits are that small after that much investment.

Isn't this the primary problem?

This isn't sensible. Empathy is feeling the other one's consciousness and qualia, and not wanting another person to experience sth you wouldn't want to experience yourself.

If you know you're speaking to AI, there's no reason to feel empathy (not that speaking either way affects outcomes anyway).

Can't believe I found sth I agree with WILL Stancil

I really haven't seen any criticisms about the bluntness or sycophancy - criticisms seem mostly about unpredictability of model routing, and subpar frontier model.

I did

Even Ethan Mollick has capitulated. I think GPT5 consensus is it's a disappointment.

Meta's Scale AI acquisition will flop. Alexander Wang is smart at marketing himself, but not at building good products.

Whenever you need to decide between two options and you ask Claude Sonnet:

- March 2024: 14%

- December 2024: 49%

- May 2025: 80%

This is the speed at which AI has improved for coding.

It's not crazy to think that it'll hit 95% by the end of the year.

This would mean AI would be outperforming all software engineers in the world except a handful, by the end of 2025.

this is my line: all agent work should be attributed to a human

it’s a simple rule that vastly simplifies just about any authz/authn conversation around agents

it also provides clear guidance for how an agent fits into an organization (socially speaking)

The friend answered my first two calls but then told me he'll get back to me tomorrow at 9am.

so what are the vibes on Claude Sonnet 4?

It's pretty disheartening to see people in leadership positions to not filter or add their own voice in their AI-generated LinkedIn posts.

Not because they use AI to create posts, I do that too; but when they don't filter the clear AI signals, it looks like they are not proficient users of AI.

Context window was the main reason when I used it and it single shotted a large game for me. Haven't tried last 2 weeks though.

If you see the usage of —, hashtags, and ending with "Curious, how do you..." posts on LinkedIn, you can be sure that they are the kind of AI slop posts that the author didn't even bother to add their own perspective at the most basic level.

Unpopular opinion: The "memory" feature of any agent (including the recent ChatGPT one) is extremely overrated.

Privacy concerns aside, the setup is simple, and infinite context isn't coming soon so it'll still involve a lot of "choice" of what to store.

Ok, @anthropic.com; you're up:

Gemini 2.5 Pro: $1.25/1M Tokens
GPT-4.1: $2/1M Tokens

You're still at $3/1M tokens AND with 1/8th the context limit

Chop chop.

In my LinkedIn feed:

Content managers are saying AI can't replace content managers.

Software engineers are saying AI can't replace SWEs.

Designers are saying AI can't replace designers.

Data scientists are saying AI can't replace designers.

HR are saying AI can't replace HR people.

Time-honored tradition of compiling must carry on when you're vibecoding

Missing MCP Tools

these should exist. if it does, point it out. if it doesn't, hey it could be a great open source project!

1/n

Thanks to AI, there's a huge content slop in social media.

And soon, thanks to vibecoding, there'll be a huge app slop in social media.

The problem isn't AI, it's the social media.

With a 5x context window of Sonnet, now:

- Individual people with no coding knowledge can build more sophisticated apps

- Small SaaS companies can rapidly add, test, release features

- Larger software companies can connect part of their codebase and prototype features rapidly

I and many others have vibecoded and built simple web apps with Claude Sonnet for a while now. But it hits a roadblock in larger codebases.

It seems like with Gemini 2.5 pro, I can build much more sophisticated apps.

Some use cases:

Same as humans. Extrapolating on data but in multidimensional ways.

🔥