157: Eat The Pain

Ian explains that the show is a day late because of a Laravel event and temporarily “a dollar short” because he moved its advertising system to Outro and canceled the old subscriptions. The new platform cannot yet process payments because Stripe authentication and account-connection flows are failing. He wants Outro to be authorized to charge the show’s Stripe account directly, with money going straight to the show rather than through Outro; recurring ad purchases would be managed by Outro as packs of ad slots. Even Stripe support and the AI eventually told him to contact support, so affected sponsors are getting free ads until the problem is fixed.

Ian says Stripe’s documentation, developer interface, test URLs, and support process have become incomprehensible, comparing the company to Authorize.net, which he left roughly 12 years ago. Aaron suggests Stripe may be like AWS: difficult to replace once it is running, but Ian argues that a bank such as Chase could challenge it by investing billions, using existing payment infrastructure, and undercutting Stripe’s percentage fees. Aaron replies that large companies usually negotiate Stripe pricing and may not switch after onboarding. Ian still thinks a well-funded competitor could make the economics compelling, while admitting he would personally trust established big technology companies over a new startup payment processor.

With only minutes before recording, Ian used an ultra-fast OpenAI mode called “Plaid”—described as roughly six times faster—to move two ads into Outro. It completed the job almost instantly, although he stopped it from running extra tests, and the run cost about $400. Aaron notes that comparable fast modes may depend on subscription tiers such as Max 500. Ian says Claude gave users three free uses of its new model, which he found appealing, but neither host currently needs more than his Claude 200 and OpenAI 200 plans.

Ian asked Fable 5.2 to review a project written by an earlier, weaker model and moved his experimental “brain” system from Google Drive to Notion. Google Drive was slow, weak at handling some data types, poor for viewing individual pages, and unable to edit Markdown well; Notion’s database and mobile app better fit the system’s search and filtering needs through MCP. The migration worked but did not feel meaningfully better than Fable 5.5. Ian also tested Fable 5.2 by asking for a James Webb Space Telescope diagram for his daughter: the first version was visually weak and inaccurate, while a heavier second prompt produced a somewhat better result.

Aaron describes spending an entire weekend at a kindergarten campout at a YMCA facility on Possum Kingdom Lake, about two hours west of Dallas. He initially dreaded meeting roughly 40 other fathers and sleeping with children and dads from the school, but found the group unusually welcoming; one father even brought a large smoker for ribs and chicken. The children rode horses, used BB guns, practiced archery, hiked, and rode a pontoon boat. Aaron’s shy son Simon was invited into games by classmates and cried because he did not want to leave, convincing Aaron that the uncomfortable weekend was worth it.

Aaron says a major Laravel launch that had been expected in November is being pushed to early December. The team is pausing for an internal audit after recent security incidents, including WorkOS being compromised through an internal administrative application. They are reviewing playbooks, processes, and internal systems before putting the product in front of more customers. Aaron distinguishes Laravel Cloud from Forge: with Forge, customers ultimately own their servers, while Cloud accepts more responsibility for the infrastructure and therefore must “eat the pain” of security. Ian agrees that the current wave of attacks makes hardening and re-hardening unavoidable.

Ian argues that the security environment weakens the case for building custom internal tools: even if a company avoids major SaaS honeypots, automated scanners and supply-chain attacks can discover neglected applications. Aaron counters that deploying a small internal tool on a hardened platform may still reduce exposure compared with concentrating everything in a major service. They agree platforms can handle servers and perhaps inject fixes for widely used, severe package vulnerabilities, but application-level flaws, unpublished vulnerabilities, and the risk of breaking customer applications leave developers responsible for much of their own security. AI security tools might eventually help, but neither treats them as a current solution.

The hosts react to Caleb Porzio’s apparent plan to release Flux for React. They say Flux has been especially valuable in the Laravel and Livewire world, but a React version could reach a market many times larger, including developers who do not know Livewire. Aaron expects users to pay for it separately, even if they already own Flux for Livewire. They acknowledge that React is a crowded market dominated by tools such as Shad and Cloudflare’s UI offerings, but argue that Caleb has an existing base of Laravel users, custom integrated components, documentation, and support. Ian says a polished, consistent UI kit can avoid the accessibility and design inconsistencies common in AI-generated interfaces.

Ian and Aaron debate why today’s developer celebrities often seem tied to companies rather than independent conference circuits. They contrast Wes Bos and Kent C. Dodds with practitioners such as Dillon Mulroy, who moved from Vercel to Cloudflare, and Sunil, who went from PartyKit to Cloudflare and then Anthropic. Aaron values their firsthand battle stories from building real infrastructure, while Ian dislikes how company changes become part of the personal brand. They also discuss Vicente Martí’s progression through GitHub, PlanetScale, and Cursor. Both see another route for package authors: extract their domain expertise and recreate it for Rails, Ruby, or another ecosystem instead of simply translating code mechanically.

Discussing recent elections, Aaron says AI labs hurt themselves by telling voters that AI would take their jobs, use their resources, or even kill them. Ian calls this an example of being “hoisted by their own petard.” They think data centers could win local support if companies paid market rates, created construction and maintenance work, funded teachers, improved parks and infrastructure, and reduced local electricity or property-tax burdens instead of relying mainly on tax breaks. Aaron also argues that AI companies should emphasize concrete benefits—such as helping researchers sequence cancer or solve scientific problems—rather than presenting an optimistic example like booking a vacation. They expect data centers to continue expanding but want more realistic, constructive communication.