Agentic Coding's Hidden Costs: Database Regressions and a 20-Month Build Bloat

Key Takeaways

1. Arvid Kahl argues performance regression benchmarks should become standard for agentic AI, which can silently lock up production databases.
2. He traced a Laravel deploy slowdown to 105,861 accumulated compiled view files, cutting build time from 7m22s to 53s.
3. Rasty's coding-agent-built iOS app went from $66 in month one to roughly $2,000–3,000 per day, with lifetime purchases driving most revenue.
4. Sam Parr reports creators with under 100K followers earning over $1M in annual ad revenue, almost entirely from AI companies on Instagram.
5. Alex Lieberman says Muse hit 10.9M downloads since September 8, while most users tap only a fraction of its capabilities.

1. Agentic AI and Engineering Discipline

1. Arvid Kahl argues that "performance regression benchmarks" should become as common as unit tests, because agentic AI will modify and "optimize" code without understanding production database behavior or load patterns, and a single change can lock up a production database even when development and tests show no problems. He recommends tests that explicitly execute queries, run EXPLAIN ANALYZE, and check query planning logic, ideally against production-scale data, though fixed test data is acceptable. His standing rule for any agent touching a database is to run read-only test queries against a protected production read replica, then write a failing test and iterate on the query until it passes and runs fast; he credits this with avoiding dozens of regressions — via 1
2. Kahl used Claude to diagnose a gradually slowing Laravel deployment, with the main bottleneck in npm build: Claude proposed hypotheses and verification commands, which Kahl executed himself on the server, leading to a fix that cut production deploys from 10 minutes to 2 minutes. The root cause was Laravel Jetstream's bundled Tailwind config scanning storage/framework/views/*.php; zero-downtime deploys create new compiled view directories on each release, but storage is shared across versions and never cleaned, accumulating 105,861 files and 2 GB over 20 months, which Tailwind read on every build (twice, due to symlinks), causing builds of over 7 minutes, 20–30 GB of memory use, and strain on co-located Redis. The fix was deleting one line from tailwind.config.js and adding php artisan view:cache to deployments, taking the build step from 7m22s to 53s; he advises Jetstream users on zero-downtime setups like Envoyer or Forge with shared storage to check their Tailwind config — via 1

2. Small Teams, Agent-Built Products, and Distribution

1. Rasty built his first iOS app with coding agents in about three weeks; it earned just $66 in its first month with few downloads and now generates roughly $2,000–3,000 per day. About 60% of purchases are lifetime deals, which account for over 80% of revenue — via 1
2. Noah Kagan shared AppSumo Radar sales data for LeadOcean, run by a self-funded founder from Nepal: $3,819 on day 1, $84 on day 2, $3,371 on day 10, totaling $20,560 from 120 customers over 11 days. His conclusion: building products has gotten cheaper, but customer acquisition and distribution have not — via 1
3. Ben Tossell shared a reusable operating workflow for a personal agent, including a video and a post with copy-paste instructions; the referenced tweet says the @bot can be invoked in comments, has email built on @RhysSullivan's Executor, and now includes a custom domain, expiration, host, and linked local repositories as a single entry point linking self-built mini apps — via 1

3. AI Spending, Audience Building, and Product Adoption

1. Sam Parr says he spoke last week with four creators each earning over $1M in annual ad revenue: nearly all their ads come from AI companies, they have fewer than 100K followers, and most ads run on Instagram. He also knows three people earning over $10M a year with only part-time assistants or a few freelancers. In his view, AI companies that raise funding must spend marketing budgets, creating a "trader gold rush" he expects to last at least a few more years; many brands have approached him about Twitter ads, which he tends to decline, preferring LinkedIn or Instagram if he likes the product — via 1
2. Alex Lieberman says Muse has reached 10.9M downloads since launching September 8, but most people use only a small part of its capabilities. He had heavy user @JJEnglert demonstrate uses from easy to advanced: merging all calendars, scanning Facebook for important contacts and adding birthdays, recommending gifts two weeks ahead and buying them after confirmation, learning streaming and cinema tastes to generate a scrollable watchlist and book tickets, finding a shared free evening for a couple each month and completing the booking, and even booking online or calling via its built-in browser. After photographing all clothing, Muse can do daily outfit checks, trip packing lists, and weekly recommendations for pieces to fill gaps; connected to a $30 ESP32 device with microphone and touchscreen, a build that previously cost hundreds of dollars was completed by Claude in 20–30 minutes after Muse released its SDK — via 1
3. Alex Lieberman distilled 11 audience-building principles from growing Morning Brew into a company valued at $75M with tens of millions of followers: content done right is a compounding asset, so spend 80% of effort on one polished piece and 20% on repeated distribution; the best distribution business is an IP business; and distribution is attention asset management across rented (social), owned (email/podcast/community), and monetizable audiences. He argues AI should be used heavily for distribution but must be invisible to end users — if it's noticeable, either the AI use is lazy or taste is lacking. Curation matters more than ever as creation costs approach zero and supply grows exponentially while attention stays finite; media companies are good at generating attention but bad at harvesting it, while non-media companies are the reverse — via 1

4. Founder Strategy and Market Signals

1. acquire.com forwarded the view that one of the best times to sell a company is when there is no reason to sell: the business is growing, customers are happy, annual profit is $500K, and cash is ample. At that point a $4M offer can be negotiated seriously, bargained hard, or unfavorable terms rejected, and a failed deal still leaves a profitable business; founders who wait until burnout, slowing growth, or urgency to exit usually face harder negotiations — via 1
2. Flippa added 72 First Access deals this week and opened 67 to all buyers, including a 25-year-old upcycled fashion brand at $250K (FA), an 8-year-old SVG animation SaaS at $1.1M, a YouTube channel with 2.18M subscribers at $902K (FA), and a multi-channel lash brand at $500K — via 1
3. Alex Hormozi argues life success depends less on the number of wins and losses than on how big you win when you do, so make bold bets and you only need to be right once; he also frames overexpansion as a talent problem rather than a scale problem, and distinguishes the 0-to-1 stage, which is most fun, from 1-to-N, where the money is — via 1 2 3
4. Codie Sanchez argues starting a business is easier than ever, but so is consuming business content all day without acting; she lists willingness to be rejected, willingness to look clumsy for years, and outlasting everyone as success factors — via 1 2
5. Nick Huber says founders should first master management, marketing, sales, hiring, delegation, and leadership, citing his own experience in his 20s doing "unscalable" grunt work like moving boxes for college students across states and renting warehouse space, which built six figures in savings by age 26 before he built his first self-storage facility. He advocates simple, direct, money-making businesses to practice fundamentals rather than starting with talk of "scale" and "moats" — via 1