Factory Fires Board Observer Over Alleged Cognition Contacts as AI Coding Rivalry Escalates

Key Takeaways

1. Factory says it fired Chris Degnan, citing undisclosed talks with Cognition executives while he served as board observer and advisor; Factory also claims rapid growth and says Cognition lags on cost, quality, and safety.
2. Tony Dinh tested ChatGPT Pro 500's Ultrafast mode and found it consumed 20% of the weekly quota in about 1.5 hours, calling it not worth it and saying he will return to Opus 5.5.
3. Tw93 detailed a Mole icon overhaul using Phosphor Icon for about 70% of icons and hand-drawn SVG for the rest, then revised several icons after user feedback.
4. @levelsio highlighted a reported decline in at-home hosting and described organizing 70 weekly co-working days plus other social events to counter age-related loneliness.
5. Jon Yongfook argued that "build in public" is dead as a strategy, that distribution can be bought, and that he may stop announcing new businesses on X until roughly $500K MRR.
6. Arvid Kahl said his podcast data platform Podscan was acquired by Audiohook, describing the deal as closer to a merger than an exit.
7. Jack Ellis outlined a 2026 database migration method that replays six months of API requests against old and new models to verify identical output.

1. AI Coding and Model Competition

1. Factory said it dismissed Chris Degnan for unethical behavior, alleging that while serving as a Factory board observer and advisor he had formal and repeated private exchanges with executives at competitor Cognition, violating confidentiality obligations. Factory also said demand for its model-agnostic coding agent and software factory products is strong, with a 10x larger team and 100x year-over-year revenue growth, and claimed Cognition lags on cost, quality, and safety while having sent engineers to extract information under the guise of interviews. These are Factory's claims as relayed in the post, and the underlying dispute is disputed/unverified — via 1
2. Tony Dinh said ChatGPT for Business is poor value: lower quota floors, fewer features, higher price, and no Ultrafast mode, so he canceled that account and switched to a personal one. He also tested Ultrafast on the Pro 500 plan and found it burned about 20% of the weekly quota in roughly 1.5 hours, calling it not worth it and saying he would go back to Opus 5.5 — via 1 2

2. Product Craft and AI-Assisted Work

1. Tw93 upgraded Mole's icons by choosing Phosphor Icon over Remix Icon, using about 70% directly and drawing the remaining 30% as SVG with Opus 5.5 for consistency. He advised testing icon changes inside the real product alongside sibling icons, and stressed alignment, consistent sizing, vertical centering, and font-icon alignment for both Chinese and English text. He also argued products should have a sense of "spirit," citing a coffee cup for Caffeine, a wave for hello, and a stethoscope for diagnostics, and said AI helps coding and baseline aesthetics but leaves humans more important for polishing products so 1% of users feel instant recognition — via 1
2. After user feedback, Tw93 revised the icons again under the principle that icons should signify rather than decorate: check-for-updates became a clockwise arrow, GPU became a dual-fan graphics card, archive became a box, audio changed from a musical note to headphones, clean screen returned to a system brush, about returned to ⓘ, and memory returned to a chip, with some icons changed three times — via 1
3. Jack Ellis described his 2026 database refactor method: dashboards are easy to verify but APIs are harder, so because he stores all API request history he can replay every API request from the past six months against both old and new data models and verify identical output. He also said he mostly agrees with DHH's AI views but laughed until his head hurt at one related item, and that after only a few days he no longer watches videos that are obviously generated by Opus 5.5 because they look alike, advising creators to invest real creativity and a distinct energy — via 1 2 3

3. Founder Strategy and Business Models

1. Jon Yongfook questioned AI restaurant booking as a startup direction, saying founders may be in a first-world bubble in tier-one cities and lack awareness of real problems. He said he often finds Jev use cases in large-scale coding and noted a direct competitor appeared less than a month after launch, arguing there is no moat. He rejected X's fixation on "distribution," saying distribution can be bought through ads, relevant forums, and sponsored YouTube channels, while what X calls distribution is really earned media that takes time but is not the only path. He said he may stop announcing new businesses on X until roughly $500K MRR — via 1 2 3 4
2. Jon Yongfook argued that "build in public" is essentially marketing: past sharing brought traffic, word of mouth, and signups, with an unspoken pact among non-overlapping projects and a period of technical lead. Now, he said, the technical lead is gone, the pact is gone, and software is oversupplied, so that era is dead; the current play is to find spaces where developers are not saturated, technology can provide value, and buyers are non-technical, while staying low-profile. He also said a new SaaS reached a certain MRR three days after launch, with the name and figure withheld, and cited a company hit by AI and delisted from the NYSE as a lesson for those who think data is a moat — via 1 2 3
3. Arvid Kahl said his podcast data platform Podscan was acquired by Audiohook, and he will stay on to lead Podscan into its next phase. He described the deal as closer to a merger than an exit, noting Podscan's technology already powered part of Audiohook's platform and that a customer becoming the acquirer brought existing trust and alignment — via 1

4. Social Life, Loneliness, and Platform Dynamics

1. @levelsio highlighted Derek Thompson's article "You Are No Longer Invited to Dinner," which reported that the share of Americans hosting friends or family at home at least monthly fell from 42% in 1975 to 12% in 2026, about a 70% drop, based on DDB Needham surveys used by Putnam and a new replication by Data For Progress. The article ruled out two comforting explanations — that dinners moved to restaurants, since solo dining is rising and face-to-face socializing is falling overall, and that other social activities replaced gatherings, since exercise and travel increased but exercise is mostly solo and time at home has risen sharply since the 2010s. It offered four explanations: busy dual-income households, intensive parenting, shrinking friend networks, and homes that are too comfortable, and closed with a "Lump of Leisure" theory framing the decline as a coordination failure rather than a personal problem — via 1
2. @levelsio said he actively fights the age-related loneliness that becomes almost inevitable after 30, citing research that people see friends less after 30 and men especially have almost no social life in old age, but that it is preventable and requires effort and initiative. He has organized 70 weekly co-working days with work, show and tell, and mutual help, recently including hardware, plus movie nights, art nights, barbecues, a World Cup final projected in the garden, and collage nights with cheap gossip magazines, colored paper, and glue where about four hours passed with no one looking at their phone. Attendance ranged from 20 to 2, and he and @rameerez still try to organize weekly; he said a small group usually organizes most activities, that no one organizing does not mean no one wants to attend, and that starting with a co-working day — bring a laptop, provide WiFi and coffee — is a good first step, adding that active socializing matters as much as or more than health and fitness and that they barely drink, relying on food, sparkling water, or kombucha and winding down naturally after a few hours — via 1
3. @levelsio said X's AI replies have been worse than ever in recent weeks, especially in the first few minutes after posting, when he is flooded by AI responses. He also said he once kicked an alcoholic out of his friend group because the person posed a fatal risk to the group, after multiple warnings, arguing alcoholics do not realize friends will gradually leave because they have become a burden; he cited the person shouting "I LOVE ALCOHOL" with arms wide in a normal restaurant, buying shots for all the staff, and forcing them to drink when they did not want to, and said cutting ties was the right decision — via 1 2
4. Tibo said he analyzed 292,784 TikTok videos and found AI videos averaged 1.4 million views versus 661,400 for real videos, meaning AI videos got 2.1x the views, and advised not ignoring AI video. He also said there is no need to commit to Claude or Codex because the best model changes frequently, recommending an agent platform that can switch between them without losing context, and said he is building such a product. A forwarded post described a user considering testing Outrank on their site, with an existing SEO stack mainly OpenSEO plus Claude/Codex and some custom skills for research, content, and monitoring, who thought Outrank might further automate the workflow but was unsure whether its output would beat their own setup — via 1 2 3
5. Arvid Kahl shared a Matic founding team essay arguing the home robotics industry has the order wrong and that Matic chose to do "eyes" (perception) before "hands" (manipulation): its visual SLAM reached 99.999% across more than 16,000 homes, is the only scaled, purely visual, fully autonomous unsupervised home robot, and has covered 500 million square feet and 400,000 miles, with the next phase focused on manipulation. In the quote post, the account said that after using Matic it is unlike any vacuum it has owned and called it the best vacuum it has used, emphasizing it is not an ad, and praised the founding team for sharing openly — via 1