Chirag Mehta

What a stellar founding team! Data sovereignty is foundational to business sovereignty which is becoming increasingly important as geopolitical situation continues to get worse.

Add this to the fast-food CEO job description: demonstrated ability to take an absurdly large bite of a burger.

There's also an early evidence on teams seeing much better results when they use LLM from the beginning for new things they build versus partially fixing old things because LLM tends to capture the "working context" of people as opposed to inferring arbitrary things from reading an old code base.

They absolutely do just like anyone who writes code without an LLM. The feedback that I am hearing is that instead of focusing on which method might result into more debt, people are looking at containment (and remedial) to leverage LLM-driven acceleration.

They are very intentional about not generating something they can't replace.

For new code, they have taken an approach of containment. Instead of using LLMs everywhere, they are generating new code for specific products. That helps them measure how well they are doing with "new potential tech debt" and maintenability in general.

A couple of teams I spoke to rewrote their old code (debt) by generating new code. They wouldn't have resources to prioritize this without an LLM. It's in production for a year and according to them their debt is down and overall code quality has improved.

Has the percentage traffic from social media materially changed for Techdirt in the last few years? For many folks, LinkedIn and LLMs have boosted traffic.

Not necessarily defending their decision but sometimes forcing folks to use and report issues is a much faster way to "catch up." There are also optics issues - AWS can't really ask their customers to use Kiro in good faith if they themselves aren't using it.

Google gets a flak for being monopolistic but it's very hard to build products for the mass audience. It requires maniacal focus, patience, sticking to a super boring process, and a lot of money.

Going forward, one of the most important assets for an organization would be a digital knowledge-base of "what good looks like." Transferring institutional knowledge that experienced folks hold in their heads to what new hires need would be a pre-req to effectively use AI.

Yes, I do know that. In the US, laws vary, but generally speaking, not seriously enforced.
My question was more about perceived and real safety of autonomous vehicles. You did answer my question. Thank you! if you're willing I have a follow-up: how about safety of other cars?

I'm not for or against anything but curious to understand your perspective. With regards to jaywalking (NOT at lights whether lights are green or not), in your view, would driverless cars be less, equal, or more dangerous to people crossing, compared to cars with a human driver?

If I don't understand how it works it must be insecure.

We are at an airport restaurant. A robot just drove up to us with our food, said, "Hi! Here I am!" and then drove away with our food.

people who don't already understand the value of testing will generate a lot more bad software. people who know when and how to test are already doing victory laps

For one of these ice breakers, once I took ice cubes from our office kitchen and broke them into small pieces. My teammates were thrilled but the moderator wasn't.

Yes....but also 😂https://bsky.app/profile/chrislhayes.bsky.social/post/3mdgx4veshk2u

Best thing since Vegemite

What's the UK IKEA version of a $1.50 hot dog that IKEA sells in the US?

It feels really bad when we are in it but I have faith in humanity to adopt and adapt. I wasn't there in the workforce for the early paper to computer automation which drove profitability but since the 80s productivity gains haven't gone up. It feels like we are in another cycle and we will adapt.

Interesting....has this changed the way you construct your phases or is it natural for you to complete them under 200k with compaction off?

I met James Lindenbaum in 2009 and Salesforce acquired Heroku in 2010. As Salesforce "sunsets" Heroku, here's a blog post that I wrote in 2010 cloudcomputing.blogspot.com/2010/12/sale...

2b. Do you see AI training and generation as more analogous to the ‘search engine/fair use’ fights of the past, or is music different enough that we need new doctrine? What’s the most overlooked distinction?

2a. As AI makes ‘in the style of’ and near-derivative music feel like piracy to many artists, what protections do you think are actually justified, and which are likely to backfire? If ‘copyright’ isn’t the right lever, what is?

1b. If a model’s behavior is shaped by both third-party data and the provider’s tuning, ranking, and product design, what’s the right test for when the provider is acting more like a ‘publisher’ versus a product maker?

Fantastic! I have a couple to ask.
1a. Where do you think Section 230 should apply to generative AI, and where should it clearly not apply, once you separate (i) liability for AI outputs triggered by user prompts from (ii) liability tied to training and data collection?

Ok, which one sounds better?

A) It’s an old little amazing house.
B) It’s an amazing little old house.

B, right?

Which of these?
1) She was wearing a huge green raincoat.
2) She was wearing a green huge raincoat.

You chose 1, didn't you?

WHY??

"Ken...*told* USA Today." I am pretty sure it must be more emotional than "told" like yelled, exclaimed, exasperated..

MCP is a great example. A couple folks at Anthropic began using it internally before publishing it. People complained it doesn't do this and that but developers loved it and started building on it. This is all in less than a year. It's not about simple/complex but how useful and universal it is.