Husky Claims 4.5x MLX Speed as Tenex Launches Enterprise AI Agents and Jacky Chou Rebuilds to $250K/Month

Key Takeaways

1. Husky, a model-specific inference engine, is claimed to reach up to 4.5x Apple MLX speed and 730 tokens/sec on MacBook with Underdog's Pareto-frontier models.
2. Tenex launched Tenex Agents, enterprise-ready AI agents for finance, sales, marketing, HR, and IT, targeting deployment in under two weeks.
3. Jacky Chou rebuilt to 13 revenue streams totaling $250K/month after a Google algorithm and AI Overviews drop from $500K/month, with 80% of LocalRank's launch revenue from one 1,800-view YouTube video.
4. Ken left Amazon and Meta after 12 years, built the text idle RPG Harpagia in 6 weeks with AI tools, and reached $155K in first-year revenue with $500/month costs and zero ad spend.
5. Whatnot reached a $20B valuation in 6.5 years, with cumulative sales growing from $2.3M in 2020 to $1B two years later.
6. Tally Forms reached $6M ARR fully bootstrapped with a 10-person team and over 2.5M users.
7. Ryan Levesque, 44, sold Ask Method in 2024 and bought a 150-acre Vermont farm for about $5M, with roughly $35M in liquid assets and $500K annual farm costs.

1. AI Models and Infrastructure

1. Husky is described as a model-specific inference (MSI) engine that claims up to 4.5x the speed of Apple MLX, with Underdog's Pareto-frontier models reaching up to 730 tokens/sec on a MacBook; the post says local models now combine speed and capability and can be tried as a personal private AI via a linked page. 1
2. Tenex launched Tenex Agents, enterprise-ready AI agents aimed at repetitive work in finance, sales, marketing, HR, and IT, with a target deployment time of under two weeks. The product emphasizes oversight through permissions, approvals, and audit trails; a performance optimization layer covering model routing, evaluation, and context management; configuration for enterprise processes and systems of record; and binding to agreed outcomes. Tenex says it has built hundreds of custom solutions for large global companies and that Tenex Agents is an 18-month, composable suite tested in customer workflows. It cites surveys with hundreds of executives naming three priorities: faster results at lower cost, clarity on which workflows to start with and how to measure success, and fit with existing systems plus access control and human oversight. Tenex Agents is rolling out with an intent form. 1
3. YouSpot was introduced as an AI-native single-person CRM for solo businesses, side hustlers, and personal-brand creators, positioned as a second brain for managing customer data and supporting growth, developed by @dharmesh and the HubSpot Next team. 1

2. Founder and Operator Playbooks

1. Jacky Chou fell from $500K/month after Google algorithm updates and AI Overviews, then rebuilt 13 revenue streams now totaling $250K/month across an SEO agency, SaaS, and e-commerce brands; the main project Indexsy brings in $85K/month and LocalRank reached $20K MRR at launch. His approach is diversification plus a bet on YouTube: 80% of LocalRank's launch revenue came from a single video with 1,800 views, while a Twitter post with 30K impressions drove 20%. His YouTube playbook is to pick a niche, upload daily, teach solutions, ride early trends, use hooks, and drive to an email list. 1
2. Ken left Amazon and Meta after 12 years and used AI tools to build the text idle RPG Harpagia for RuneScape players in 6 weeks; it earned $155K in its first year, averaging $13K/month, with over 50K iOS and Android downloads, a 6% conversion rate versus the 2–5% industry average, $3 revenue per active user, $500/month costs, and zero ad spend. He set a condition to leave Meta if the game naturally reached $10K/month; six months after launch he posted on Reddit with a screenshot and free explanation, and despite harsh comments revenue jumped from about $1,000 to $10,000 almost overnight, so he left that month. His conclusion is that you do not need millions of players: a few hundred active users in a niche can produce over $100K/year. 1
3. Tally Forms reached $6M ARR while fully bootstrapped, purely product-driven, with a 10-person team and over 2.5M users, and shared lessons and mistakes from growing from $5M to $6M ARR. 1
4. Ryan Levesque, 44, ran the information and software company Ask Method for over a decade with annual revenue above $100M, reinvested profits, sold the company in 2024, and bought a 150-acre Vermont farm for about $5M. He now has roughly $35M in liquid assets across public markets and VC/PE funds, while the farm costs about $500K/year in total, including about $220K in mortgage and taxes plus $175K–$260K in operations, livestock, equipment, and labor. The farm produces all the family's protein, raises cows, pigs, chickens, and turkeys, and has 48 beehives and over 1,000 maple taps. He said: "I don't know if this is ever going to be enough, but it's certainly enough right now"; "Wherever you are, the number you want is about 2x, and that doesn't change"; and "If I doubled my net worth tomorrow, what would I do? I can do all of those things right now." 1

3. Market Signals and Deal Flow

1. Whatnot's founder took the company to a $20B valuation in 6.5 years; the author notes this exceeds the combined market caps of Wendy's, Under Armour, Hertz, Harley-Davidson, and American Airlines, and says that in the AI boom such numbers are thrown around casually but "this is just too much." Whatnot's yearly sales grew from $2.3M cumulative by the end of 2020 to $163M the next year and $1B the year after. The author calls it a good reframing of the idea of a "billion-dollar company." 1 2 3
2. Acquire.com listed several startups for sale: an e-commerce coffee subscription with $135K annual revenue, $21.5K profit, 11% growth, asking $69.9K; a B2B founder media brand newsletter with 45K subscribers and over 20K organic LinkedIn followers, $12.5K annual revenue, asking $53.2K; an AI job search and auto-apply SaaS with over 20K users and over 1.6M job listings, $42.2K annual revenue, asking $150K; and an AI text humanizer SaaS with $74K annual revenue, asking $49K. The account also shared the view that founders should calculate the risk of not selling: customers may churn, competitors may appear suddenly, technology may change quickly, growth may slow or churn may rise, and the founder may get tired of running it; keeping the company is as much a bet on the future as selling, and sometimes cashing out millions now is better than betting another 3–5 years on everything going well. It added that even a good business can lose momentum if the listing leaves buyers with too many questions, and advised assessing a listing's strengths and areas for improvement before going live. 1 2 3 4 5 6
3. Flippa shared M&A expert Matt Bradbury's point that selling during a three-year growth peak can command premium valuations, while flat or declining revenue significantly damages valuation. A YouTube comedy channel with over 174K subscribers was listed for sale with $34.4K monthly profit, 99% margin, 320 original comedy videos, AdSense monetization, and an audience spanning Gen Z and millennials. 1 2

4. Tools, Pricing, and Platform Friction

1. Noah Kagan proposed an "internet law" that SaaS products raising prices by more than 20% should be required to allow one-click export of all data; he cited Evernote doubling its price and Confluence raising prices, and said he migrated to Outline and saves $8,000 per year. He also said he is considering how to pay someone responsible for his software product $1M and asked others about the highest salary they have paid a single employee. 1 2
2. @levelsio criticized Google's laptop naming as confusing, noting three completely different product lines—Pixelbook, Chromebook, and Googlebook—and suggested renaming to Chromebook Pro; the quoted post announced Googlebook's official launch, saying it combines ChromeOS and Android and targets Android phone users. The account also shared criticism of Microsoft's product naming and login experience: multiple different URLs, including www and /chat variants, correspond to personal Copilot apps that can log into different accounts, while multiple enterprise Copilot URLs are the same app; after trying to log out, revisiting still shows a logged-in state, which was described as "you can't log out of Microsoft," with the response that this is treated as a feature rather than a security issue. 1 2
3. @levelsio praised the WIP platform, where creators share daily completed tasks and the platform analyzes tools mentioned in those tasks to determine whether people are evaluating, adopting, using, or abandoning them, along with tool sentiment and commonly paired tools; it also surfaces trending tools automatically, such as Jev ranking first with its icon and description fetched by an enrichment agent. The goal is to help creators choose tools and help tool makers understand user preferences and churn points. The account also described a personal health data system: logging food in Telegram for calories and protein, WHOOP for training and sleep, weight from a scale, bedroom air quality, temperature, and humidity from Xiaomi into Home Assistant, step and other health data auto-POSTed hourly from iOS Health Auto Export to a server, travel data from the @nomadscom API, and sauna use from the @wip API, all to find correlations with health. It called on X to make "reduce AI replies" settings persistent, saying they reset on web and iOS with every next post, causing X to be flooded with AI replies again. 1 2 3
4. Patrick McKenzie described an air traffic control delay: the airline rebooked him for the next day and issued a new boarding pass but refused a hotel voucher, saying the company does not issue vouchers for delays caused by air traffic control. 1