AI Code Review Becomes the Norm as Builders Rethink Moats and Workflows

Key Takeaways

1. Arvid Kahl argues AI makes public codebases easy to audit, so developers should scan and fix vulnerabilities before others find them; closed source carries less exposure, though frontend code and API calls remain entry points 1.
2. Kahl says dependency upgrades can largely be done with one agentic AI prompt, but only with detailed instructions: upgrade major versions one at a time, consult migration docs, keep tests green, and check visuals each round 1.
3. @levelsio is pursuing remaining AI-era moats through paid city-data APIs, a $1 community, X revenue around $15K/month, and 187,279 blog and Telegram subscribers combined 1 2.
4. Noah Kagan replaced constant task inflow with a weekly Claude review of customer feedback, ranking issues by affected users and reviewing only the top five fixes himself 1.
5. Ben Tossell spent a full day building and debugging an agent setup in @get_bb_app, with threads for coding work, email updates, and a shared task list and work log 1 2.
6. A Starter Story founder at $18K/month says Reddit largely stopped working after five or six posts led to a ban, while SEO now drives growth across about 200 blog posts and 30 landing pages 1.
7. acquire.com listings show small AI SaaS businesses for sale, including a code-generation tool at $63.2K TTM revenue asking $65K and a photo editor at $20.4K TTM asking $57.3K 1 2.

1. AI Changes Code Review and Upgrade Work

1. Arvid Kahl says AI makes it easy for anyone to audit public codebases for vulnerabilities, so developers must run their own scans and fix issues before someone else finds and exploits them. He adds that closed source carries less risk, but frontend code and its API calls can still serve as an entry point for investigation 1. He also describes a shift in attitudes toward AI reviewing pull requests: laughable last year, uncertain this year, and likely normal next year 2.
2. Kahl says dependency upgrades can mostly be handled with a single agentic AI prompt, provided the instructions are detailed: upgrade major versions one by one, consult upgrade docs and migration blogs, ensure other dependencies remain compatible and tests pass, do a visual check each round, and ask before proceeding when something is incompatible. He says the more verbose instructions work and the task still completes in one pass 1.

2. Builders Rethink Moats, Agents, and Weekly Systems

1. @levelsio says he is chasing the remaining AI-era moats much as he chased the AI wave in late 2022. His listed moves include paid city-data APIs priced at $2,490 to $9,990 a year, a $1 community membership to expand network effects, about $15K a month in X revenue, and 187,279 combined blog-email and Telegram subscribers 1 2. Other investments include eight offline meetups in 30 days, hotel-booking improvements with a “longevity hotel” filter, an annual vibe-coding event with about 1,000 participants and more than 1 million players this year, a 1990s retro-PC mini replica, and about $1K a month in merchandise 3 4. On capital, he says he holds $SPCX stock from Cursor's acquisition and plans to keep it, bought the IB01 ETF as a hedge against an AI bubble, and is considering more land and coastal property; he also questions which startups remain worth backing once large AI and tech companies absorb everything, calls robotics interesting but easy to enter, and says he avoids regulatory-license moats and would sell parts of projects to raise investable capital 5 6.
2. Noah Kagan says family illness, more than 10 days of poor sleep, and frequent AI product launches left him overwhelmed, and he concluded that few things truly matter and constant distraction was blocking effective work. He moved all-day task inflow to a once-a-week process and uses Claude to summarize the past seven days of customer feedback weekly: grouping by problem, ranking by number of affected users, drafting a suggested fix and reply for each, and leaving him to review only the top five items 1. The same loop compares his predictions for X post views with actual results, identifies why forecasts were wrong, updates the rules, and suggests five posts for the coming week; he cites one case where the same data with a different headline went from 7,000 to 27,000 views and says he is applying the loop to AppSumo's SEO 2.
3. Ben Tossell says he spent a full day building and debugging his agent configuration: @bot now opens threads in @get_bb_app for formal coding work, sends bot updates by email, and updates a shared task list and work log. He describes the setup as a “vibe-slopped superhuman + todos” Linear-style system where @bot can also be tagged in comments; the email runs on @RhysSullivan's Executor and includes domain, expiry, host, and linked local repositories, with plans to make it a one-stop entry point linked to his own small apps 1 2.

3. Distribution and Small SaaS Economics

1. A Starter Story founder at $18K a month says Reddit has largely stopped working: after five or six posts he was banned, and real growth now comes from SEO. He has about 200 blog posts, roughly 30 different landing pages, and a site localized in four or five languages 1.
2. acquire.com listings show small AI SaaS businesses on the market: an AI code-generation SaaS that turns screenshots into shippable web code reports $63.2K TTM revenue and a $65K asking price, while an AI photo editor for iPhone and Android reports $20.4K TTM revenue and a $57.3K asking price 1 2. The account also observes that each market cycle seems to favor a category, with agencies currently in favor, though deals continue across categories and different businesses give buyers different reasons to enter 3.

4. Product and Tooling Notes

1. Tony Dinh says he likes Ultrafast and uses GPT 6.1 Sol and Astra through it, arguing the speed lets him return to a flow state without frequent multitasking and context switching 1. Alex Lieberman praises Google Maps for labeling which subway car to board to exit fastest, calling it the kind of small, clever feature he favors and suggesting it reflects either firsthand experience of the post-exit crowd or strong empathy for subway riders 2.
2. Arvid Kahl says breakthrough tools make already capable people stronger while those without the underlying ability may gain less 1. Patrick McKenzie argues ADHD, or the set of personality traits often non-medically called that, tends to be adaptive in parts of the modern economy and maladaptive in more traditional organizations, and he speculates AI may amplify that effect 2.
3. Ben Tossell relayed that WebGPU support is coming to uzu with an early preview available, that Delegations is generally available so teams can hand work from existing tools to Droid, and that a Claude Code-related project has been open-sourced with fixes for copy-paste, links, and other interactions 1 2 3. He also relayed a Supabase Select badge contest running through the end of the month, with participants breaking badges, building projects, and sharing with #SelectBadge for Supabase credits and free tickets to next year's event 4.