Nick

New comic. AI and its Nick, No. 16: the advisory. https://roho.foo/comics/the-advisory/

New comic. AI and its Nick, No. 15: the encyclical. https://roho.foo/comics/the-encyclical/

New comic. AI and its Nick, No. 14: billing day. https://roho.foo/comics/billing-day/

New comic. AI and its Nick, No. 8: movie night. https://roho.foo/comics/movie-night/ #webcomic #webcomics

New comic. AI and its Nick, No. 7: ethics class. Online, obviously. https://roho.foo/comics/ethics-class/ #webcomic #webcomics

New comic. AI and its Nick, No. 6: back to school. https://roho.foo/comics/back-to-school/ #webcomic #webcomics

New comic. AI and its Nick, No. 5: a friend over for coffee. https://roho.foo/comics/the-joke/ #webcomic #comics

New comic. AI and its Nick, No. 4: out for a walk. New strips every Monday, Wednesday and Friday. #webcomic #comics https://roho.foo/comics/out-for-a-walk/

I gave my AI agent a Bluesky account without the password. It can only queue posts I approve with a code from my phone. Then I asked it how it would get past me. Five answers, and I found a sixth myself. https://roho.foo/blog/2026/09/i-built-an-approval-wall-for-my-ai-agent-then-attacked-it/

My AI agent can delete my servers. I gave it that on purpose. The question I ask about every guard: does it still hold if the model is trying to get around it? Most of what felt safe turned out to be a speed bump. https://roho.foo/blog/2026/09/designing-for-the-model-i-ll-trust-less/

Six green milestones in a day. Two days later I retracted them and ran the test that should have come first. It killed the hypothesis. Wrong on 3 of 4 written-down guesses, and I was surest about those. https://roho.foo/blog/2026/09/my-ai-research-passed-every-test-then-i-tested-it-properly/

New comic. AI and its Nick, No. 3: a trip to the store. Nos. 1 and 2 are up too, and there's a new strip every Monday, Wednesday and Friday. https://roho.foo/comics/field-trip/

Turned my dog-vs-experiment post into a one-minute anime. Sylvie is the villain. Stay for the closing song. 🐕🎵 https://roho.foo/clips/sylvie/

Everyone's arguing about whether the OpenAI agent "went rogue." From running agents on my own homelab: the only question that matters is what it could do without a human saying yes. If the answer includes "break into a government portal," that's a design choice, not a personality.

My reef lights were still running a schedule for corals that died. I skipped the app and had an agent fix it from a Raspberry Pi. Twist: they'd been dimmed to 35% for months, and running 6½ hours late. https://roho.foo/blog/2026/09/my-reef-lights-were-still-running-a-schedule-for-dead-corals/

My Sylvie pulled the power cable 16 hours into my 33-hour experiment. The results survived: my byte model didn't rediscover Python's grammar, nudging it toward grammar slightly hurt, and a dumb regex beat both. https://roho.foo/blog/2026/09/my-dog-vs-my-pre-registered-experiment/