Key Takeaways
- ReachAgents pitches ads inside AI agents, while NVIDIA's Factory harness runs Nemotron post-training in an airgapped network.
- A critical NGINX HTTP/2 flaw prompts an auto-update reminder, and a 3,728-account spam list targets automated AI replies.
- AppSumo's transparency test barely moved conversion but cut refunds 16%, and a screen-time app scaled from $10K to $400K monthly.
- Small SaaS exits and a 103-item AI use-case list show where founders are applying agents and structured queries.
1. AI Agents and Infrastructure
- Ben Tossell's feed promoted ReachAgents, which lets advertisers send full brand and product information to users' AI agents so the agents can surface personalized ads in-feed when user intent arises, rather than following a GEO approach 1. The same feed amplified a claim that the platform is the only agent platform to withstand a true airgap, is modular and model-agnostic, and that maintaining frontier capability under those constraints is hard but worthwhile 2. Separately, the feed relayed that NVIDIA used Factory to post-train Nemotron entirely inside its own airgapped network, describing Factory as the only commercially supported harness NVIDIA deployed in that environment and noting it enables long-running agents and full execution traces 3.
- Dharmesh argued that LLMs are powerful but questions answerable by deterministic structured queries should still go through queries, rather than having a model read 8 million Slack messages and emails to infer revenue; as models converge, cost and latency become the competitive edge for software companies 1. He also highlighted YouSpot, which tracks unanswered emails like a check-engine light and generates draft replies for users to approve, edit, or ignore 2. In a "HubSpot re-founding" day-6 note, he said the hard part is deciding which old practices to keep and which to replace, recommending teams drop prior assumptions and ask what they would do without past constraints and if they only solved the customer problem 3.
2. Security, Spam, and Platform Operations
- @levelsio called the NGINX vulnerability "scary" but said his VPS had already auto-upgraded; he urged enabling automatic upgrades and restarts, acknowledging the risk of unattended restarts but prioritizing staying current and secure, and cited a post describing two severe NGINX memory-safety flaws (CVE-2026-123633 and CVE-2026-123007) that allow unauthenticated remote code execution when HTTP/2 is enabled 1. He also published an "𝕏 SPAMMERS LIST" of 3,728 accounts repeatedly using automated AI replies to farm engagement, with 498 already banned by X; the list is continuously updated and offers evidence per blocked account, one-click blocking, a JSON API, and a plain-text username list, built on Clankwall and a prior list, and he said he submitted it to X CEO @petersellis to train spam models 2. He added that some people seem to want the AI-reply bot problem to continue 3.
- Noah Kagan reported an AppSumo 50/50 test across 187,000 visits showing live deal-page metrics (volume, refund rate, response time, last-active time) barely changed purchase conversion: 4.49% vs. 4.48%, revenue per visit $7.67 vs. $7.41, a gap too small to call; refunds fell from 9.06% to 7.60%, a 16% drop. He concluded full transparency did not produce more purchases but helped more buyers keep what they bought 1.
3. Small-Business Growth and Product Lessons
- Starter Story profiled Yoni, whose gamified screen-time app Brainrot uses a digital pet brain that decays as users open addictive apps and FaceTimes them when they try Instagram; before launch he had built 100K–150K Instagram followers by sharing indie-hacker stories, earned about $10K on day one, and after a summer 2025 launch grew monthly revenue from about $10K to about $400K with millions of downloads, mainly via annual subscriptions 1. His scaling playbook: analyze data to find high-activity users and push others toward them, double down on proven influencer content and paid ads, bring in co-founder Brock Anderson as a "business amplifier," and keep running small onboarding experiments—nine in ten fail, but one success can change the trajectory 2. His product advice is to build a simple utility that solves the core user's core problem, and his top lesson is that many ways exist to feel busy without producing anything, so identify changes worth 20%, 50%, or 100% of effort and delegate or drop the rest 3.
- Starter Story also covered Kris, a 25-year software engineer and part-time CTO who launched peptide-tracking app PeptIQ without a novel idea—dozens of competitors already existed—and without paid ads, influencers, or virality; he entered based on GLP-1 tracker success and a steep Google Trends rise, built an MVP in about a week, and waited six to seven months for SEO to work 1. PeptIQ now gets about 70% of clicks and revenue from Google, its blog drove 590K site clicks over 12 months, and this year revenue exceeded $115K with 1,400 active subscriptions at $30/month or $130/year behind a hard paywall; his conclusion is that you do not need a new idea, but a growth channel others call dead 2. A third profile covered Milly Tamade, who runs Generalist World from a 200-year-old house without central heating on a Scottish island of 191 people, generating $30K/month; she lists "staying calm" as her first KPI rather than revenue, writes in the morning, takes meetings and partnerships in the afternoon, stops at 15:00, and sleeps at 21:00, and over four years the business has earned nearly $600K across a newsletter, community, job board, summer school, and self-created festival, with no investors and a small team of friends; her largest partnership, a one-year deal with Beehiiv, came from showing up with completed work rather than waiting to be chosen 3.
- Nick Huber described a roughly $75M self-storage deal in 2021 that he said could have created tens of millions in enterprise value for Bolt Storage but was won by a competitor 1. Using a pet-waste cleanup company with over $600K revenue, $300K profit, and five crews as an example, he noted its signup form had confusing wording and unclear pricing, and that an outsider testing it clicked the wrong place—showing owners cannot rely on explaining the form themselves 2. His growth method is to have someone unfamiliar with the business walk through the customer journey without help to see whether they understand the service, see pricing, and complete signup 3. He also argued about 95% of businesses do not truly want employee innovation, mentioning it only in mission statements while expecting people to follow orders and avoid trouble, and that owners and layers of management with high profits lack incentive to take more risk; he advised choosing ambitious owners and companies in chaotic change, and if an employer will not give that opportunity, to look elsewhere 4.
4. Exits, Use Cases, and Market Signals
- Acquire.com listed several small startups for sale: an AI RAID-log and risk-management SaaS for PMO teams at $15K ARR/TTM asking $37.5K 1; a Shopify app suite for average order value at $19.2K TTM asking $59K 2; a Shopify AI SEO and content app with $360K TTM profit, $462.8K TTM revenue, and 1,100 paying stores asking $1.2M 3; and an AI eBay listing tool at $20.5K TTM asking $24K 4. It also relayed that founders often underestimate the financial risk of having their entire net worth in one company, and that selling lets them diversify and escape monthly revenue pressure, citing a founder who netted about $3M 5, and argued valuations should reference actual buyer transaction data rather than sounding reasonable, offering a free valuation service 6.
- Alex Lieberman shared a list of 103 AI use cases that were impossible two years ago, sorted by marketing, content, competitive intelligence, product, design, sales, management, hiring, health, family, consumer, social, learning, and real estate. Marketing examples include scraping competitors' Meta and LinkedIn ad libraries daily to log start and stop dates and identify long-running winners, monitoring competitor status pages to trigger branded ads during outages, and asking ChatGPT, Claude, and Perplexity 500 buyer questions nightly to track mentions and citations 1. Product examples include running 10 simulated users through the product nightly and auto-filing bugs with repro steps and recordings, tagging feature requests from support tickets, sales calls, and churn records by customer and revenue to calculate the dollar value of each request, and scanning public GitHub repos using the company SDK to summarize actual use cases 2. Competitive-intelligence examples include tracking competitor GitHub stars and npm and Hugging Face downloads daily, downloading competitors' public JavaScript nightly and diffing it to spot unreleased features from new flag names like enable_agent_mode_beta, and checking competitors' and target customers' public DNS records weekly to infer newly signed vendors from entries like openai-domain-verification 3. Sales examples include building a prototype live on a call from the buyer's described need, matching recent customer references by industry, size, and use case, listening for seat, module, and timeline details to generate a quote from the price book before the call ends, and identifying new stakeholders when a proposal is opened by someone the seller has not met 4. Lieberman also said he built a skill that identifies any company's AI stack by scanning eight source types including DNS records, vendor case-study walls, and job descriptions, and that testing it on Disney surfaced tools including Claude Enterprise, Cursor, Adobe Firefly, Databricks, ElevenLabs, and Notion, useful for sales research, investment diligence, competitive intelligence, and AI-readiness benchmarking 5.
- Flippa relayed Brad Roth telling Steve McGarry that a defensive growth strategy may have cost him at exit; despite 98% customer retention and healthy cash flow, he called it a mistake 1. Codie Sanchez said running a business long term comes down to three things: do not run out of cash, check the bank account weekly, and treat every customer as if no next one is coming 2, and argued people should not work harder but be more decisive about what to do 3. Justin Welsh said he increasingly respects people who get rich and fade from public view 4 5, relayed that he has seen smart but broke people talk about business while less capable people make $80K/month 6, argued that once self-employed, being directed by another adult feels absurd 7, said ambitious people's biggest mistake is caring too much about strangers' opinions and too little about their own 8, and defined solo entrepreneurship as becoming "unemployable": free enough, picky enough, and allergic enough to nonsense that no job can compete with the business built for oneself 9.
