AI Coding's Next Layer, Personal Software, and the Founder Traits Behind Unicorns

Key Takeaways

1. Arvid Kahl argues AI coding should move to higher-level, intent-driven languages, and that giving coding agents read-only production database access makes implementation and testing more effective.
2. @levelsio warns the AI job transition may be a 10–20 year hard period, while his browser-based retro computing project added over 100,000 users in two days.
3. @levelsio also argues software is entering a "personal software" era where AI chat apps generate bespoke tools, and the opportunity is to become tools that agents use and charge.
4. Justin Welsh and dharmesh highlight founder traits and business model diversity, including speed, direction, mission, advice integration, and early hiring as load-bearing decisions.
5. acquire.com listings show small SaaS, newsletter, and service businesses with concrete revenue and asking prices, plus a case for taking $3M now over chasing $10M later.
6. Codie Sanchez shares lessons from meeting billionaires and launching a product, including questioning whether a system should exist at all and the gap between audience and customers.

1. AI Coding and Agent Workflows

1. Arvid Kahl argues that AI coding should move toward higher-level languages rather than using AI as a reason to go lower-level. He envisions expressions like "persist p50 and p95 values hourly for this dataset" or "verify the HMAC signature of this response," with humans acting as architects rather than bricklayers. The cited view holds that the goal is high-level language with low-level performance, that human understanding is necessary for every runnable software system, and that the current bottleneck is higher-level languages optimized for human understanding. 1
2. Kahl also recommends giving coding agents a read-only user for the production database when conditions allow, preferably a read-only replica. When an agent can safely inspect and verify production data, implementation and testing become noticeably more effective. 1

2. Personal Software and the AI Transition

1. @levelsio argues the transition period of AI replacing jobs is underestimated. The claim that "AI won't replace jobs, it will create jobs" ignores the phase where old jobs are replaced first and new ones have not yet appeared. The Industrial Revolution transition took 50 to 80 years, and today's pace is faster, so the AI era could mean a difficult 10 to 20 years for many people. 1
2. @levelsio says his nostalgia project added over 100,000 users in two days. The project ports classic systems and games such as Windows 3.1, Windows XP, and Quake to run in the browser via WASM, and uses WebSockets for multiplayer interaction. About 80,000 people experienced it yesterday. He positions it as a non-commercial interactive "online computer museum" and explicitly says it is not for profit. 1
3. @levelsio proposes that "software is dead, entering the era of personal software." Ordinary people will not say "make me software"; they will state specific needs, such as manually retouching photos or viewing quarterly sales across e-commerce platforms. AI chat apps will directly generate a Photoshop clone or an analytics dashboard that connects APIs and includes scheduled tasks, and users may not even realize they are building software. He believes most people will mainly use personalized, temporary software inside AI chat apps. Moats such as social media communities and e-commerce platforms like Shopify and Stripe remain, but they will become tools for AI-built personal software, so the opportunity is to become tools that agents use and to charge agents. 1 2
4. @levelsio calls this a "bubble within a bubble," arguing the tech community still misjudges whether ordinary people are already generating personal software, and rebuts the view that ordinary people will not use it because they do not understand technology. On the "it's not time yet" judgment, he agrees with the long-term direction but opposes using it as a business argument, saying exponential AI progress will quickly break such judgments and that people should prepare for and accept a "post-software" reality. The cited view emphasizes that most people still do not use AI, most of those do not pay, and most only use it as search, so it is still very early. 1
5. On whether software is merely functional, the cited view distinguishes functional software such as PDF readers from experiential software, which is like books: people still want to experience works into which others have poured care. @levelsio responds that this is paying for creativity, hence an AI-assisted creativity explosion, while spreadsheet, analytics, and CRM applications may not be creative. A forwarded item mentions a person without a computer science background using browser-based Claude Code to write tools for a business, expanding from a simple reporting app to an ERP and canceling some software subscriptions, but that person believes people who can conceive and visualize what to build from scratch are rarer than imagined. 1 2

3. Founder Traits, Business Models, and Exits

1. Justin Welsh forwards the view that people often mock those who change their positions over time, but the real concern is people who never evolve. He also observes that the wealthiest people he knows drive ordinary cars and live in ordinary houses, yet have unusually large amounts of discretionary time. 1 2
2. dharmesh forwards the claim that there is no single correct path in business models: an SEO consultant earning $1.5 million a year with few clients and no employees; a Skool community founder making seven figures teaching AI; an agency with nine-figure revenue where the owner is still an individual contributor; and a founder like @dharmesh running an 11-figure market cap public company with no direct reports. He also forwards that Shopify's internal atmosphere is "very magical" and that this is the best period in tech industry history, hoping others feel similarly. 1 2
3. dharmesh forwards his summary after investing in enough seed-stage unicorns: without polishing out the common traits of unicorn founders, it is hard to build a company worth over $1 billion. Trait one is speed made of pace: the best founders have a physiological discomfort with slow progress, constantly restructure the path to goals, find clever shortcuts, and rethink roadmaps from first principles; speed creates high momentum and compounds knowledge and product, otherwise faster competitors may overtake within a year or two. Trait two is speed made of direction: the best founders regularly step back from execution to judge whether the path is still correct, whether there is a better route, or whether the goal itself has changed; this requires intellectual honesty and the willingness to cut plans that no longer make sense, and the best founders set quarterly recurring calendars to reassess all goals. Other traits include a mission that must go beyond financial results and be true to oneself, otherwise founders lose motivation and are overtaken by hungrier competitors; proactively seeking advice from smart people and immediately integrating good advice, reserving weekly time to learn from others and building a world-class advisory board; and early hiring "brick by brick," treating every early employee as a load-bearing structure, because the first 10 employees' foundation determines the ceiling, a single bad hire in a key role compounds, and once the right people are hired the team itself becomes a competitive moat. 1
4. Julian Shapiro, based on his own seed-stage investing experience, summarizes five traits shared by unicorn founders: a sense of speed (instinctive discomfort with slow progress, continuously restructuring paths, rethinking roadmaps from first principles to maintain first-mover advantage), a sense of direction (regularly stepping back from execution to judge whether the path and goal are still correct, requiring intellectual honesty and the courage to cut plans that no longer make sense, with a recommendation to set a fixed quarterly review), mission identity (a sincere mission beyond financial return, otherwise losing motivation and being overtaken by hungrier competitors), proactively seeking advice and immediately integrating good suggestions (reserving weekly learning time, building a world-class advisory board and selling them on the mission), and brick-by-brick hiring (the first 10 employees are load-bearing structures that determine the subsequent ceiling, requiring finding the world's best person for each role rather than hiring nearby from one's network, because early key-role mis-hires compound). 1
5. Alex Hormozi offers a method for deciding whether to keep a leader: imagine what would happen if they left. If the answer is "nothing would happen," that is probably the leader's actual current contribution. 1

4. Acquisitions, Small Businesses, and Lessons from Operators

1. acquire.com lists a SaaS review-generation software for law firms that collects Google reviews and referrals, with TTM revenue of $217,000 and an asking price of $985,000. It also lists an online baby shower platform with over 400,000 guests and 13,000+ paying families since 2020, TTM revenue of $38,900, and an asking price of $49,000. A CFO-themed newsletter with 68,000+ subscribers, TTM advertising revenue of $20,000, a 56% profit margin, TTM total revenue of $20,500, and an asking price of $55,000 is also listed. A high-end publishing and ghostwriting agency that helps professionals become published authors is listed with TTM revenue of $1 million and an asking price of $1.2 million. 1 2 3 4
2. acquire.com forwards the view that taking $3 million today may be better than chasing a $10 million exit in five years. Within five years, a large competitor may appear, AI may change the industry, growth may slow, or the founder may simply tire of running the business, while $3 million in cash can fund the next venture or serve as a successful outcome. It also says that even with all business metrics in hand, a founder may not know what the company is worth, which is normal; having real exit data as a starting point makes valuation easier, and it offers a free rough valuation based on what buyers actually paid. 1 2
3. Codie Sanchez summarizes 12 lessons from a two-week trip, saying she met three billionaires, launched "Own or Be Owned," brought 170,000 people through 12 profit levers, met YouTube executives, recorded podcasts, and launched a new business tech product. She argues against optimizing a broken system and instead questions why it exists: citing Joe Liemandt's Alpha School, where children complete core academics in two hours a day and reportedly learn twice as much, most people would use AI to improve a six-hour school day by 10%, while Joe asks why six hours are needed. She says Jenny Just turned $1.5 million into a multi-billion-dollar company, PEAK6, with an edge from nearly 30 years of decisions with immediate feedback and real risk, and that confidence is a credential earned by repeatedly experiencing risk. She cites David Adelman's expansion from student housing Campus Apartments, with over $2 billion in assets, into alternative assets, family offices, and the Philadelphia 76ers, and suggests thinking about which assets one can own and make other assets more valuable, rather than only asking what business to build. She says the billionaires she met are unusually comfortable being misunderstood, make fewer but bigger decisions, and go all in. She also says an audience is not a business: speaking to 170,000 people about 12 profit levers over three days only matters if they convert into customers. She says a live launch exposed weaknesses in messaging, technology, team, offer, and follow-up: Zoom could not handle the volume, payments failed, and Amazon delisted a product because demand looked fake. She says information is cheap, most owners already know the problems, and they only need to act; the best operators converge a meeting into one constraint. 1
4. Alex Lieberman forwards that he claims to use AI more than 99% of the world, has tried personal agents such as Muse, OpenClaw, and Grokbot, and is starting a personal agent series. Part one documents three weeks of actual experience with one agent, covering what worked, what failed, and predictions for the future. He also forwards a weekend consumption list: a Jayshetty interview with Tony Robbins on mindset, a David Perell interview with Morgan Housel on nonfiction writing, a Dave Chappelle stand-up special, a David Senra interview with Ari Emanuel on building WME/TKO, and Gary Halbert's "The Boron Letters" on copywriting. He says that amid accelerating technology and rising pressure to learn AI, he tries to spend equal time studying how people work, narrative and creative thinking, and trust building. 1 2