Chris Mairs

2025 saw big gains in LLMs, with smaller models now very capable. But human-level embodied AI is still years away and AGI likely a decade off. Assistive AI hasn’t progressed much yet, but I’m optimistic and excited to announce an Assistive AI hackathon at Entrepreneurs First on Jan 17.

Browser accessibility is almost always a nightmare and even the world’s most popular websites like Amazon are a total pig to navigate without sight due to the overwhelming amount of content. New Computer Use Agents could be a game changer for blind users

I just created my first website, The Blind Mans Arms, without code or help from a sighted person. It’s simple, but shows how easy it is to spin up the digital side of a business today. It even has a conversational voice agent. Running a great gastro pub, though, still takes real human skill.

Humanity’s Last Exam is a fiendishly hard benchmark for human like intelligence. But a good score at this benchmark does not correlate to simple tasks in the real world like making tea. Assistive tech for independent living in a visual world must handle such mundane tasks robustly.

Audio Narration0:00/410.3053061224491× A key measure of AI progress is the length of time a model can work on so called ‘long horizon tasks’ without going off the rails. Most high value real world jobs involve some ‘long horizon tasks’, i.e. tasks that take hours rather than

My life is extraordinarily good—I feel lucky and hardly notice my blindness, though I love calling it an affliction. I can accept bad infrastructure, but not lazy digital inaccessibility. Conversational voice AI can finally repay the world’s vast inclusivity debt.

LLMs can be brilliant yet bafflingly bad. Often it’s faster, cheaper, and more reliable to just write code—or better still, skip the chat entirely and do things the old-fashioned way.

AI replacing junior roles risks creating unstable, inverted corporate pyramids unless leaders and policymakers act to preserve diamond-shaped structures in industries with traditional “up-or-out” cultures.

Effective use of a Large Language Model requires effective prompting. Anthropic’s new Skills feature lets enterprises and users provide structured context that Claude loads dynamically, making prompting more reliable, consistent, and performant.