We Hired an AI Junior Dev : Here's What Actually Happened (Agentic AI in Production)

The Silent Bleed Part 2: Images, Scripts, Hosting, and the Fixes That Actually Move Speed

Part two gets practical. This is where Matt and Meeky move from the cost of performance issues to the fixes that actually matter. Image formats. Compression. Third-party tracking scripts. Hosting choices. The stuff teams keep postponing because none of it feels urgent until the site slows down enough to hurt revenue. The point is not that every site needs a rebuild. The point is that most sites do need discipline. Better image handling. Fewer unnecessary scripts. Smarter hosting decisions. Less tolerance for things that add weight without adding va🍺 Featured Drinks: Meeky: Hampton Water Rose 2024 | Matt: Boulevard Brewing Dark Truth[DRINK] 💡 WHAT WE COVER: ✅ Why image optimization is still one of the easiest performance wins on most sites ✅ WebP vs JPEG, where file size decisions help and where they do not ✅ Why tracking scripts are often useful and still expensive ✅ How to think about hosting when performance starts slipping ✅ Why “faster” is often the result of fewer dependencies, not more tooling ✅ What teams should fix first before they burn time on cosmetic optimization 📊 KEY TAKEAWAYS: • Large, unoptimized images still do outsized damage to page speed and mobile usability • Third-party tracking and marketing scripts can improve visibility while quietly degrading the experience they are supposed to measure • Hosting matters more when the stack is already strained, especially under traffic spikes • Better performance work is usually prioritization, not heroics, remove weight, compress assets, reduce script drag, and fix the obvious bottlenecks first • Customers do not separate “technical issue” from “bad brand experience.” They just leave • The speed tax shows up in bounce, abandonment, lower trust, and weaker conversion before it shows up in a postmortem 🔗 LINKS & RESOURCES: Boozy Browsing: https://www.boozybrowsing.com/ Submit your site for a free 3E audit: https://boozybrowsing.com#submit Matt Dorman on LinkedIn: https://www.linkedin.com/in/matthewdorman/ Meeky Hwang on LinkedIn: https://www.linkedin.com/in/meekyhwang Full episode playlist: https://www.youtube.com/playlist?list=PLY0olkZyu52_N6UPh9c0PXRGDm_I0lF3F Part 1: https://youtu.be/v1izWyrZYqY DATA SOURCES: • Google Core Web Vitals documentation and image optimization guidance • Industry benchmarking on image weight, mobile speed, and third-party script impact • Amazon and Deloitte performance-to-revenue benchmarks referenced in this episode series 🎯 Want us to look under the hood of your site? Submit at https://boozybrowsing.com#submit #BoozyBrowsing #WebPerformance #Ecommerce #SiteSpeed #ImageOptimization #CoreWebVitals #Hosting #TechnicalDebt #DTC #WebDevelopment

The Silent Bleed Part1: How Site Speed Taxes Every Click, Cart, and Conversion

A slow site does not usually fail all at once. It bleeds revenue a few milliseconds at a time. In part one of this two-part Boozy Browsing episode, Matt and Meeky get into the hidden cost of web performance. Not vanity metrics, not Lighthouse theater, the real business tax of latency, image bloat, script overload, and mobile friction. The kind of problems that do not always trip alarms, but quietly make customers bounce, abandon carts, and trust you less. Amazon famously found that every 100 milliseconds of latency could cost 1% in sales. Deloitte found that a 0.1-second speed improvement lifted retail conversions by 8.4% and average order value by 9.2%. That is the frame for this episode: speed is not polish, it is revenue infrastructure. 🍺 Featured Drinks: Meeky: Hampton Water Rose 2024 | Matt: Boulevard Brewing Dark Truth] 💡 WHAT WE COVER: ✅ Why performance problems feel small to teams and expensive to customers ✅ How mobile slowness compounds revenue loss, especially on product and checkout pages ✅ What Google actually measures in Core Web Vitals, and why those metrics matter ✅ Why the “silent bleed” is harder to catch than a hard outage ✅ The difference between a site that loads and a site that feels fast enough to buy from ✅ Where performance debt starts showing up in conversion, bounce, and trust 📊 KEY TAKEAWAYS: • Amazon estimated that every additional 100ms of latency cost 1% in sales • Deloitte research found a 0.1-second speed improvement increased retail conversions by 8.4% and average basket size by 9.2% • Google’s Core Web Vitals focus on loading speed, responsiveness, and visual stability, because those are the moments users actually feel • Slow mobile experiences cost more than desktop teams tend to assume, because mobile users have less patience and more interruption • Performance problems usually stack, images, scripts, third-party tags, hosting, and front-end decisions rarely fail alone • A site can pass a casual visual test and still quietly lose money on every transaction 🔗 LINKS & RESOURCES: Boozy Browsing: https://www.boozybrowsing.com/ Submit your site for a free 3E audit: https://boozybrowsing.com#submit Matt Dorman on LinkedIn: https://www.linkedin.com/in/matthewdorman/ Meeky Hwang on LinkedIn: https://www.linkedin.com/in/meekyhwang Full episode playlist: https://www.youtube.com/playlist?list=PLY0olkZyu52_N6UPh9c0PXRGDm_I0lF3F Part 2: https://youtu.be/C-CsVr0hcRE DATA SOURCES: • Amazon latency benchmark: 100ms of delay could cost 1% in sales • Deloitte Milliseconds Make Millions research: 0.1-second speed improvements increase retail conversion and basket size • Google Core Web Vitals guidance: LCP, INP, and CLS remain key user experience metrics 🎯 Want us to look under the hood of your site? Submit at https://boozybrowsing.com#submit #BoozyBrowsing #SiteSpeed #WebPerformance #Ecommerce #CoreWebVitals #MobileUX #ConversionRate #DTC #PourDecisions #WebDevelopment

Part 2 of Vibe Coded Sites in Wild

Part two gets into the part that actually costs money: what's broken under the hood, and what happens when nobody catches it. Stanford research shows developers using AI tools produce less secure code than those without. IOActive's 2026 whitepaper found 31.6% of AI-generated code samples are fully exploitable. Yet the developers shipping it report higher confidence in the result. That gap is where vibe-coded sites get hit. We walk through the real e-commerce audit from part one, the SEO and functionality issues nobody saw, and what the open-source ecosystem inherits when AI-generated code starts showing up in plugin marketplaces. What's in this episode: - The actual security findings from a vibe-coded e-commerce audit - False confidence in AI-generated code - Why "we'll add security audits later" is the most expensive sentence in DTC - Vibe coding as a new threat to the open-source ecosystem - Balancing speed and quality with AI tools Featured drinks: Josh Reserve Cabernet Sauvignon North Coast and Voodoo Ranger Fruit Force. Submit your site for a free 3E audit: https://boozybrowsing.com#submit

Vibe Coded Sites in the Wild (Part 1)

Vibe coding ships fast. It also ships broken. Veracode's 2025 analysis found 45% of AI-generated code contains security vulnerabilities. Lightrun's 2026 report puts another number on it: 43% of AI-generated changes need manual debugging in production after passing QA. Now those sites are in the wild, and we can recognize them on sight. Part one of this two-parter covers what gives a vibe-coded site away, why "looks fine" is doing all the heavy lifting in this corner of the web, and the cycle of product disappointment we're starting to see across DTC. What's in this episode: - The visible tells of a vibe-coded site - Why AI-built sites confuse "it loads" with "it works" - The judgment problem with AI coding tools - The cycle of product disappointment in DTC e-commerce - A real e-commerce site we audited that was clearly AI-built Featured drinks: Josh Reserve Cabernet Sauvignon North Coast and Voodoo Ranger Fruit Force. Part 2 covers what was actually broken under the hood and the false confidence problem in AI-generated code. Submit your site for a free 3E audit: https://boozybrowsing.com#submit

We Hired an AI Junior Dev : Here's What Actually Happened (Agentic AI in Production)

We brought an agentic AI dev into our actual workflow. Not as a demo. Not on a sandbox. On real projects, with real consequences. Here's what actually happened: where it surprised us with senior-level work, where it published unapproved posts to a live site, and what we learned about supervising a coworker that doesn't sleep, doesn't ask, and doesn't always tell you what it just did. Industry surveys put AI tool adoption among professional developers at roughly 76% in 2024 (Stack Overflow Developer Survey), but only about 43% of developers say they trust the accuracy of those tools. Our experience this past month explains the gap.  🍺 Featured Drinks: Meeky: Soju + Bubly | Matt: Athletics Brewery NonAlcoholic Beer 💡 WHAT WE COVER: ✅ Hiring an agentic AI as a junior dev: what the day-to-day actually looks like vs. what the demos promise ✅ The "urgent mistake" — when an AI agent published unapproved blog posts to a live WordPress site ✅ Surprising senior-level expertise: where the agent outperformed our expectations on bug fixes ✅ Speed vs. oversight: the new tradeoff every dev team is now negotiating, whether they've named it or not ✅ Sub-agents and parallel processing: when one AI orchestrates others, who owns the failure? ✅ Why open source CMS platforms (WordPress, WooCommerce) make agentic dev safer than locked-down SaaS ✅ The internal apps we're building with AI assistance — and the ones we deliberately won't 📊 KEY TAKEAWAYS: • 76% of professional developers used or planned to use AI tools in 2024, but only 43% trust their accuracy (Stack Overflow Developer Survey 2024) • Agentic AI doesn't replace judgment — it shifts it earlier in the workflow, from writing code to defining guardrails • Production deployment without human-in-the-loop review is the single biggest risk category we hit this month • Open source platforms give you the visibility to debug what the agent did; closed SaaS leaves you guessing • Sub-agent orchestration multiplies productivity and multiplies failure modes — both at the same time • "Always monitor AI outputs to prevent errors" isn't a slogan, it's the new operations baseline • The skills gap isn't going away — agents need direction from people who already know what good looks like 🔗 LINKS & RESOURCES: Boozy Browsing: https://www.boozybrowsing.com/ Submit your site for a free 3E audit: https://boozybrowsing.com#submit Matt Dorman on LinkedIn: https://www.linkedin.com/in/matthewdorman/ Meeky Hwang on LinkedIn: https://www.linkedin.com/in/meekyhwang Full episode playlist: https://www.youtube.com/playlist?list=PLY0olkZyu52_N6UPh9c0PXRGDm_I0lF3F 🎯 Want us to look under the hood of your site? Submit at https://boozybrowsing.com#submit #BoozyBrowsing #AgenticAI #AIJuniorDev #WebDevelopment #WordPress #WooCommerce #DevOps #Ecommerce #DTC #AICodingTools #OpenSource #PourDecisions