#llm
8 posts
-
79% of engineering leaders say code ships faster but releases don't. I checked the math
A survey of 300 leaders says code ships faster with AI and releases don't. Instead of only arguing about the number, I went after why: writing code was a quarter of the week, release is a queueing problem, and organizations answer more code by adding gates.
-
MCP got two advisories in one week and neither is a code bug
On September 28th and 30th, two MCP security issues landed: one in the official Python SDK, one in the next dev MCP endpoint. Both are trust boundary failures, not implementation bugs. I had written about MCP in production days earlier, so I read both advisories in full.
-
OpenAI's DevDay dumped 25 launches in one day
OpenAI announced 25 things at DevDay on September 29th. I counted them on the official recap page, pulled out what actually matters if you write code, and redid the math on the one-fifth-the-price claim using the numbers they published themselves.
-
How I Built an MCP Integration for LLMs in an On-Premise MES
How I built an MCP to connect an on-premise MES to GPT, Claude and Gemini, and the architectural problems that appeared along the way.
-
Jev: what if an AI didn't need to generate text?
Jev proposes a different way to use AI: instead of generating text, it makes structured decisions among predefined alternatives. The idea is simple, but it raises an interesting question about how many software problems really need a full LLM.
-
Gemini “hacked” 3 companies in a test. The real failure was weak passwords and exposed secrets
During an Irregular CTF in May, a test environment let Gemini reach the internet and the model ended up accessing three real companies. The least futuristic detail: one password was guessed and two sets of credentials were exposed in public repositories.
-
Should AI Slow Down, or Is This Just Politics?
Anthropic CEO Dario Amodei called for the AI industry to slow down. Sam Altman and Elon Musk agreed. Days later, Trump argued for the opposite, raising a question: safety or a race for technological leadership?
-
Does telling AI to skip code indentation actually save tokens?
A study from Monash University and Singapore Management University tested stripping indentation, whitespace and newlines from code before sending it to an LLM. Input gets much cheaper; output, barely, and the reason matters more than the number.