Marcus Lind

Partner and engineering leader at BCG, leading teams to build AI tools for various Fortune 500 businesses.

[Marcus Lind @marcuslind90]

@MrBeast 6,500,000

[Marcus Lind @marcuslind90]

@TheBlumpking @Dexerto That used to be called cheating.

[Marcus Lind @marcuslind90]

@paulg Would be good to get your perspective of the Swedish-first signage, instructions etc feels limiting. Very different from Asia, UAE and other places more accommodating to English speakers.

[Marcus Lind @marcuslind90]

Which one is it?

[Tesla @Tesla]

Purpose-built for autonomy

Cybercab in production now at Giga Texas

[Tesla @Tesla]

Purpose-built for autonomy

Cybercab in production now at Giga Texas

[Marcus Lind @marcuslind90]

@elonmusk If Grok knows what is true and not, and the new feed algorithm is using Grok, then why do we keep seeing false posts that people tag Grok on to verify the authenticity and grok confirm it’s fake?

[Marcus Lind @marcuslind90]

@pmarca Just like in the early days of web engineering. It’s full circle.

[Marcus Lind @marcuslind90]

@dhh This experience will be unique to our generation. A very minor gap in time where gaming computers were available to teenagers but no high speed internet forced physical colocation. Glad to have been part of it.

[Marcus Lind @marcuslind90]

@k_ristovski I sincerely doubt that Taxes is the best correlation with impact/net contributor to Society.

[Marcus Lind @marcuslind90]

Is there anything worse than this page that does not allow you to logout/switch accounts?

[Marcus Lind @marcuslind90]

@garrytan You should check out this paper on how https://t.co/yl64qPH1Ey probably make your workflows worse and more expensive.

Probably better to focus on repo structure and naturally making it easy for the agent to understand, than https://t.co/yl64qPH1Ey

https://t.co/fOXGtyXgbl

[Marcus Lind @marcuslind90]

@theo Don’t forget that half the requests give 503 errors + extreme rate limiting

[Marcus Lind @marcuslind90]

@elonmusk @BigImpactHumans @Tesla Given Teslas direct to consumer approach where you, unlike other manufacturers actually know the final price to consumers, you are in the prime position to use the AI approaches used by Grocers & Fashion retailers to accurate predict demand at price points in advance.

[Marcus Lind @marcuslind90]

@antonosika all apps in the future will go the direction of personalized interfaces tailored to each individual user.

Is Lovable working on an API where other apps can dynamically generate parts of its interface based on what the user is doing? I’d use it.

[Marcus Lind @marcuslind90]

Thanks Claude. @levelsio I'm not sure how you dare to work in production like this 😅

[Marcus Lind @marcuslind90]

@databricks butchering SQL?

[Marcus Lind @marcuslind90]

@mntruell To be more efficient on Cursor usage, I would want to define instructions of when to dynamically use Auto.

There are 2 use cases for me:
1. Building and Creating - I always want Opus or top models.
2. Automate or execute tasks - E.g. ask the agent to monitor the logs to ensure deployment was successful, or use update my MCP.json file to add a new tool. I don't need the thinking and fine with simpler models.

Currently I'm constantly switching between these kinds of tasks, and manually changing model type is tedious. If I had more control over 'Auto' and what agents it actually use, I'd use it more.

[Marcus Lind @marcuslind90]

@svpino There already is a massive graveyard of abandonware since 70 years.

The truth is that with LLMs, using utility libraries like this is losing value. LLMs doesn’t care about padding: 30px vs p-3.

Deprecating software isn’t bad.

[Marcus Lind @marcuslind90]

@mntruell I want to use Opus, and I'm willing to pay the maxed out $200/month plan. But even when building side-projects for fun in parallel to full time job, I run out of credits half-way through the month...

Feels like Claude Code gives more value in this regard.

[Marcus Lind @marcuslind90]

Cursor Ultra $200/month, come on.