Agentic Web Development & What AI Can't Do

He started in tech at 13, learned everything online, moved from design to founding a web agency (Beza) and becoming deeply involved in the Webflow community. 'Allergic to the slow path', he's spent the last ~1.5 years pushing what can be done agentically.

Asked by Sara · answered by Julian Galluzzo

GPT-3/3.5 let him write custom scripts instead of relying on developers; then tools like Cursor made building ~20x easier than in Webflow. He migrated Memberstack's site off Webflow, and a popular video about it drove requests for help. Visual developers were scared by messy dev workflows, so he built Ship Studio to put everything in one workflow without proprietary lock-in.

Asked by Nik · answered by Julian Galluzzo

LLMs are prediction engines: for logic, calculation and predictable tasks they beat humans. Good work often means breaking from predictions using context and intuition, which they can't do. Engineers' routine tasks are largely handled by agents; designers less so, because setting direction isn't prediction.

Asked by Sara · answered by Julian Galluzzo

It comes down to articulation: if you can't articulate what you want, you can't tell if the AI did a good job. You need to know what a good outcome looks like to manage either a person or an agent.

Asked by Sara · answered by Julian Galluzzo

Nik: it's the era of operators — experts who know what good output looks like benefit most, e.g., content leads and editors as content costs fall. Sara: people overuse AI for trivial tasks (like sorting Search Console data) that a spreadsheet filter handles, adding cost and hallucination risk.

Asked by Nik · answered by Nik & Sara

It's as ready as the people managing it, because AI can't take accountability. He migrated a site with ~150 static pages, ~20 collections and ~3,000 CMS items (20–30K monthly visits); traffic rose above previous levels instead of the usual post-migration dip. His team also built a site agentically for a company that raised a $107M seed round.

Asked by Nik · answered by Julian Galluzzo

Agents can handle defined tasks like structured data, headings and alt text, so every site will have solid technical basics. SEO has always been about beating the competition, which requires taking risks and chances — something AI can't do.

Answered by Julian Galluzzo

It's about stakes. On a new site, mistakes don't matter much; at an enterprise, even tiny mistakes are costly. You can use AI for 100% of the work but must review everything it did. Sara: like a calculator, you own the output.

Asked by Sara · answered by Julian Galluzzo & Sara

Ship Studio is an open-source place for web professionals to build; features come from what builders need, and contributions (e.g., a discoverability tool) can be merged. He keeps no financial stake in it so the roadmap isn't hijacked. Specialists will remain, but disciplines will bleed into each other.

Asked by Nik · answered by Julian Galluzzo

Security above all — people paste API tokens into page settings on any platform. 'Ask not what AI can do, ask what you can do empowered by AI': if you don't know what a canonical tag is, you'll never know whether yours are right.

Asked by Nik · answered by Julian Galluzzo