Tag: AI Agents
-
What the AI Labs Aren’t Telling You About Safety
AI labs in 2026 are not telling you the full story about their safety work. They are not telling you because the full story is not flattering, because the full story would invite regulatory scrutiny, and because the full story would slow down the product release. The patterns in what the labs are not saying…
-
What “Human in the Loop” Actually Means (And Why Most AI Products Get It Wrong)
A human is not meaningfully in the loop because a product displays an approval button. Real oversight requires context, time, authority, and evidence. Many products include the human only to absorb liability.
-
Why Your Code Review Process Is Failing
The code review process that worked when the team was 5 people, writing 50 lines a day, does not work when the team is 50 people, writing 5000 lines a day. Here is what is breaking, and what works.
-
Voice Cloning and the Death of Voice Verification
Voice verification sat as the security control the bank, the call center, the enterprise helpdesk relied on for years. The five second voice sample, the my voice is my password pitch, the security control that worked until the voice cloning tools became good enough to defeat it.
-
What AI Cannot Do, Yet
A field guide to what AI cannot do in 2026, with the four areas where the models are still bad, the four areas where the models are getting better, and the part where the honest answer is that nobody knows yet.
-
When AI Hallucinates in Production
The failure looks like a confident answer that is wrong in a way the system cannot detect, and the cannot detect is the part the production team has to live with.
-
The Local LLM Stack Is Real in 2026
A field guide to the local LLM stack in 2026, with what the open weights models can actually do, what the hardware requirements look like, and the part about the workloads that are going to stay in the cloud.
-
Vibe Coding and the Erosion of Code Review
Vibe coding in 2025-2026 has eroded the code review in the typical engineering org. The reviewer used to read every line of the diff, the reviewer now reads the summary, the reviewer approves the diff, the production breaks. The erosion has not been intentional, the erosion has been the byproduct of the AI generated PRs…
-
The Day AI Stopped Being the Story
The day AI stopped being the story arrived without anyone noticing. The product launch that the press no longer covered, the research paper that the analyst no longer cited, the announcement that the executive no longer read, the day the AI hype peaked and the AI started becoming the background the work happens in.
-
LLMs Are Still Mostly Plagiarism With Better Punctuation
The training data is still mostly text written by humans, the loss function is still mostly next token prediction, and the result is still mostly text that looks like the training data, rearranged, smoothed, and served with the confidence of a system that does not know it is rearranging.