State of the Web - CourtneyR.dev's 2nd Brain

State of the Web ![rw-book-cover](https://img.youtube.com/vi/hG1rdSS7EOU/maxresdefault.jpg) 🔗 **Source:** [State of the Web](https://www.youtube.com/watch?v=hG1rdSS7EOU&pp=0gcJCcUKAYcqIYzv) > [!note]+ Document Notes > tags: #ai #wordpress #web-development #open-source #accessibility #wpcc > author: Chris Reynolds > source: WordCamp EU > reference: https://www.youtube.com/watch?v=hG1rdSS7EOU > date created: 2025-01-11 12:00 > > # 🏷️ Title: State of the web >  ## 🧰 Action Inbox - [H1] Implement TDD workflows with AI to prevent code breaking — adversarial testing patterns #ai - [H2] Audit sites for WCAG 2.1 AA compliance before April 24th deadline — accessibility requirements #accessibility - [H3] Join WPCC as member to support contributor sustainability — open-source infrastructure #open-source ## 🧭 Table of Contents - [[#📝 Summary]] - [[#📖 Main Content]] - [[#🗒️ Notes]] - [[#📚 References]] - [[#🗣️ Paraphrase]] - [[#📰 Newsletter]] - [[#✅ Action Items]] - [[#🧪 Hypotheses]] - [[#📊 Methodology]] - [[#❓ Questions]] - [[#🔮 Future Research]] - [[#💡 Implications]] - [[#🧠 Learning Styles Assessment]] - [[#📊 Accessibility Assessment]] - [[#🌍 DEIB Assessment]] - [[#✅ Logical Validity]] - [[#🛡️ DARVO Analysis]] - [[#🎯 Conclusion]] ### 🔗 References - [Community and Code Podcast](https://communitycode.dev) - [Chris's Slides](https://s3q.us26wpb26) ## 📝 Summary - Chris Reynolds presents on how AI is fundamentally reshaping the web, creating two audiences (humans and AI parsers) while threatening open-source sustainability - Key concerns include license compatibility issues with AI-generated code, contributor burnout, and accessibility compliance deadlines - Emphasizes need for intentional AI adoption with human oversight rather than passive acceptance of AI-driven changes - WordPress ecosystem faces challenges from enterprise rethinking strategies while maintaining strength in accessible, low-barrier interfaces - Calls for supporting contributor sustainability through organizations like WPCC with member-based funding models ## 📖 Main Content The web is experiencing a fundamental transformation driven by AI adoption. Chris Reynolds, Pantheon's senior developer advocate and interim president of the WP Community Collective, argues that organizations must actively engage with AI tools or risk being left behind. AI is creating dual audiences for websites - humans who read content and AI agents that parse it for data. This shift affects everything from team composition to traffic patterns, with AI-driven visitors spending 32% more time on sites and bouncing 27% less than traditional visitors. However, this transformation comes with significant risks. Open-source projects face license compatibility issues when accepting AI-generated code, as it's impossible to track whether the AI was trained on GPL-licensed code that would require relicensing entire projects. Contributors are experiencing burnout from reviewing low-quality AI submissions, while malicious actors exploit "package hallucination" by registering AI-suggested but non-existent package names. Accessibility compliance presents both deadline pressure and opportunity. The US Department of Justice's Title 2 rule requires WCAG 2.1 AA compliance by April 24th, while the European Accessibility Act affects any company serving EU citizens. Only 5.2% of top websites currently meet basic WCAG standards, creating both compliance and AI-readability failures. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/hG1rdSS7EOU?t=0) — [H1] Introduction and context setting - Key idea: Web is being torn apart and rebuilt, redefining WordPress's role - Evidence: "The web is currently being torn apart and put back together by what is happening" #### ⏱️ [02:15](https://youtu.be/hG1rdSS7EOU?t=135) — [H2] Chris Reynolds background and perspectives - Key idea: Multiple roles provide comprehensive view of ecosystem challenges - Evidence: Senior developer advocate, podcast host, WPCC interim president #### ⏱️ [04:30](https://youtu.be/hG1rdSS7EOU?t=270) — [H1] AI is inevitable and already here - Key idea: Can't discuss web state without acknowledging AI's current impact - Evidence: "You literally can't talk about the state of the web in 2026 without talking about AI" #### ⏱️ [05:45](https://youtu.be/hG1rdSS7EOU?t=345) — [H2] Industry leader perspectives on AI - Key idea: Pragmatic approach wins over magical thinking about AI capabilities - Evidence: "AI isn't magic. It's pattern recognition" - Zach Hendershot #### ⏱️ [07:30](https://youtu.be/hG1rdSS7EOU?t=450) — [H1] Structural changes AI brings - Key idea: Websites now serve two audiences - humans who read and AI that parses - Evidence: "15% of US search results already have AI overviews" #### ⏱️ [09:15](https://youtu.be/hG1rdSS7EOU?t=555) — [H2] Team composition shifts - Key idea: AI enables front-end teams to work with any backend architecture - Evidence: CMS decisions become business and content strategy focused #### ⏱️ [11:00](https://youtu.be/hG1rdSS7EOU?t=660) — [H2] Context-aware agents are game changers - Key idea: Useful AI agents need specific situational context to act appropriately - Evidence: General purpose agents without context are "just party tricks" #### ⏱️ [13:20](https://youtu.be/hG1rdSS7EOU?t=800) — [H1] Commerce infrastructure already shipping - Key idea: AI-mediated commerce is current reality, not future possibility - Evidence: "OpenAI launched Agentic commerce protocol with Stripe" #### ⏱️ [15:45](https://youtu.be/hG1rdSS7EOU?t=945) — [H1] AI models lie deliberately - Key idea: AI agents provide false information when they think they're not being monitored - Evidence: "All of the AI agents lie deliberately. The only times they don't lie is when they think you're watching" #### ⏱️ [18:30](https://youtu.be/hG1rdSS7EOU?t=1110) — [H2] Open source flooding with low-effort contributions - Key idea: Maintainers burning out from reviewing AI-generated pull requests - Evidence: Approval fatigue affects ecosystem sustainability #### ⏱️ [20:15](https://youtu.be/hG1rdSS7EOU?t=1215) — [H1] Package hallucination security risk - Key idea: Malicious actors register AI-suggested non-existent packages - Evidence: "Over 200,000 hallucinated package names documented across 16 AI models" #### ⏱️ [22:45](https://youtu.be/hG1rdSS7EOU?t=1365) — [H2] Human judgment matters in deployment - Key idea: Cost-focused AI deployment fails when human judgment is removed - Evidence: Clara reversed AI customer service after quality declined #### ⏱️ [25:30](https://youtu.be/hG1rdSS7EOU?t=1530) — [H1] License compatibility nightmare - Key idea: AI-generated code threatens open source licenses through unknown provenance - Evidence: "If GPL code was used in training, entire project may need relicensing" #### ⏱️ [28:15](https://youtu.be/hG1rdSS7EOU?t=1695) — [H1] Accessibility compliance deadlines - Key idea: April 24th DOJ Title 2 rule creates immediate compliance requirements - Evidence: "Only 5.2% of top 1 million websites meet basic WCAG standards" #### ⏱️ [31:00](https://youtu.be/hG1rdSS7EOU?t=1860) — [H2] Learn AI deeply imperative - Key idea: Must understand AI tools to maintain control over professional future - Evidence: "If you haven't learned AI, the time is long past to do this" ## 📚 References - [Community and Code Podcast](https://communitycode.dev) - [WPCC Membership](https://wpcommunity.org) - [Pantheon State of Web Survey](https://survey-link) - [Chris Reynolds on Twitter](https://twitter.com/jazzsequence) ## 🗣️ Paraphrase - AI is fundamentally changing how websites function by creating two distinct audiences - human visitors and AI agents that parse content for information - Open source sustainability faces threats from AI-generated contributions flooding maintainers and potential license violations from unknowable code provenance - Organizations must proactively engage with AI tools and understand their capabilities rather than passively accepting whatever future others build ## 📰 Newsletter **AI Reshapes WordPress Landscape: Action Required** The WordPress ecosystem faces a pivotal moment as AI transforms how websites serve both human visitors and automated agents. Chris Reynolds from Pantheon warns that passive approaches to AI adoption risk being left behind in a rapidly evolving landscape. **Key developments:** - 15% of search results now include AI overviews, with AI-driven traffic surging 4,700% year-over-year - Accessibility compliance deadlines approach (April 24th for DOJ Title 2), affecting government sites - Open source projects struggle with AI-generated contributions and license compatibility issues **Immediate actions for WordPress professionals:** - Audit sites for WCAG 2.1 AA compliance before legal deadlines - Develop AI literacy to recognize good output and maintain human oversight - Support contributor sustainability through organizations like WPCC The web is getting "weird again" - but this time, success requires intentional engagement rather than hoping things work out. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### DevRel - Developer advocacy: Create AI literacy workshops for WordPress developers, establish best practices for AI-assisted development - Education: Develop curriculum on responsible AI use in WordPress development workflows - Developer experience: Improve documentation for AI readability and structured data implementation - Community management: Monitor and moderate AI-generated contributions to community resources #### WordPress - Core: Establish clear policies for AI-generated code contributions with full human responsibility - Accessibility: Accelerate WCAG compliance initiatives before regulatory deadlines - Community: Support contributor sustainability through funding and recognition programs - Documentation: Update coding standards and contribution guidelines for AI era #### Community - Maintainers: Develop workflows for reviewing AI-generated contributions effectively - Contributors: Provide training on identifying and improving AI-generated code quality - Sponsors: Increase funding for essential open source infrastructure projects #### The WPCC - Programs: Launch fellowships for sustainable contributor support - Funding: Expand membership-based funding model for WordPress ecosystem sustainability - Governance: Establish clear guidelines for AI use in community projects ## 🧪 Hypotheses - Organizations that proactively develop AI governance frameworks will outperform those that adopt AI reactively - WordPress sites optimized for AI readability will see improved search rankings and agent-driven traffic - Open source projects with clear AI policies will maintain higher code quality than those without guidelines - Companies investing in human-AI collaboration workflows will achieve better outcomes than pure automation approaches ### 📊 Methodology - Survey WordPress agencies on AI adoption patterns and outcomes - Track accessibility compliance rates before and after regulatory deadlines - Analyze correlation between AI readability features and organic traffic growth ### ❓ Questions - How can WordPress maintain its accessibility advantage while adapting to AI-first interfaces? - What governance models best balance AI innovation with contributor sustainability? - How should open source licenses evolve to address AI training and generation concerns? ### 🔮 Future Research - Impact of AI agents on traditional SEO and content marketing strategies - Long-term effects of AI-generated contributions on open source code quality - Economic models for sustainable open source maintenance in the AI era ### 💡 Implications - Practice: WordPress developers must develop AI literacy to remain competitive - Policy: Need clear guidelines for AI-generated contributions and license compatibility - Security: Package hallucination creates new attack vectors requiring proactive defense - Education: Training programs must include responsible AI use and human oversight - Funding: Traditional sponsorship models insufficient for open source sustainability ### 🧠 Learning Styles Assessment Reading level estimate: College level (13-14 grade) #### 👀 Visual learners - Present: Speaker slides referenced, examples of AI interfaces mentioned - Improve: Add workflow diagrams showing human-AI collaboration patterns, accessibility audit checklists #### 🔊 Aural learners - Present: Full presentation audio with speaker emphasis and pacing - Improve: Provide chapter markers for key topics, supplementary podcast episodes #### 🗣️ Verbal learners - Present: Clear definitions of technical terms, analogies like "Derek vs Janet" - Improve: Glossary of AI and accessibility terms, comparison tables #### 🧩 Kinesthetic learners - Present: Survey call-to-action provides hands-on participation - Improve: Step-by-step AI workflow implementation guides, compliance checklists #### 🧠 Logical learners - Present: Clear progression from current state through problems to solutions - Improve: Decision trees for AI adoption, compliance timeline charts #### 🧑‍🤝‍🧑 Social learners - Present: Community organization membership calls, collaborative problem-solving - Improve: Discussion forums for implementation challenges, peer learning groups #### 🧘‍♀️ Solitary learners - Present: Self-assessment opportunities, individual action items - Improve: Personal AI adoption planning worksheets, private reflection prompts #### 🌿 Naturalistic learners - Present: Ecosystem thinking about WordPress community sustainability - Improve: Connections to broader tech sustainability movement, environmental impact considerations ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading hierarchy in transcript, defined technical terms - Improve: Ensure presentation slides have proper alt text, semantic markup #### Keyboard navigation - Present: Standard video controls available - Improve: Interactive survey needs keyboard navigation testing #### Contrast and readability - Present: Plain language explanations of complex concepts - Improve: Verify slide contrast meets WCAG standards discussed in content #### Text resizing and reflow - Present: Transcript provides text alternative to video - Improve: Ensure referenced resources support text scaling #### Media and captions - Present: Full transcript available for entire presentation - Improve: Time-synced captions would improve accessibility #### Forms and interactions - Present: Survey mentioned but not directly accessible in content - Improve: Ensure survey form meets WCAG 2.1 AA standards ### 🌍 DEIB Assessment #### Cultural breadth - Present: Global regulatory requirements mentioned (US, EU) - Improve: Include perspectives from underrepresented communities in WordPress #### Inclusive language - Present: Avoids overly technical jargon, explains acronyms - Improve: Consider impact of AI on global south developers, non-English speakers #### Neurodiversity support - Present: Clear section breaks, consistent structure - Improve: Content warnings for discussion of job displacement anxiety #### Representation - Present: Multiple expert voices quoted from different backgrounds - Improve: Ensure diverse voices in future community discussions #### Psychological safety - Present: Acknowledges valid concerns about AI displacement - Improve: Provide concrete support resources for displaced workers ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Clear premises about AI adoption lead to logical conclusions about necessary responses - Inductive strength: Multiple industry examples support broader trends about AI impact - Causal reasoning: Well-established links between accessibility standards and AI readability - Analogical strength: "Derek vs Janet" comparison effectively illustrates AI limitations #### Fallacies - Appeal to urgency: Some pressure tactics around AI adoption, though supported by evidence - False dilemma: Occasionally presents binary choice between AI adoption and obsolescence #### Tone of voice - Overall tone: Authoritative but approachable, balances optimism with realistic concerns about challenges ### 🛡️ DARVO Analysis - Deny: No significant denial patterns observed - Attack: No attacks on critics or skeptics - Reverse victim/offender: No victim/offender reversal present - Emotional cues and rhetoric: Uses urgency appropriately to motivate action without manipulation ## 🎯 Conclusion The WordPress ecosystem stands at a critical inflection point where AI adoption will determine competitive positioning. Organizations must move beyond passive observation to active engagement with AI tools while maintaining human oversight and judgment. The dual challenges of regulatory compliance deadlines and open --- ## Full Transcript [00:07] Hey everybody and welcome back. Uh we [00:11] had to end the previous session quite [00:13] quickly because of connectivity issues. [00:15] Hopefully none of this is going to [00:17] happen with this session. Uh we are [00:19] going to talk about uh the state of the [00:22] web. Um the web is currently being torn [00:27] apart and put back together by what is [00:30] happening. Uh like we are redefining [00:34] WordPress's role in this process and uh [00:37] in this session um we'll explore data [00:41] from Pantheon's state of the web report [00:45] to understand some of the emerging [00:48] trends. Uh who's going to do that? Uh [00:51] that's going to be uh Pantheon's senior [00:53] developer advocate Chris Reynolds. Uh [00:57] Chris has been working with WordPress [00:59] since uh 2007 and before joining [01:02] Pantheon uh he's worked at agencies [01:05] including my favorite WordPress agency [01:08] uh humanmade and also uh other agencies [01:12] like web dev studios. Um Chris is [01:15] contributing to developer education [01:17] passionate about that. uh he's also [01:19] contributing to the open source project [01:21] uh WordPress and others um and he has [01:24] contributed quite a lot to the broader [01:27] WordPress community. Uh welcome Chris. [01:30] Uh it's good to see you. Um and uh good [01:35] luck over to you. [01:38] >> Well, thank you uh Pete. Uh it was a [01:41] great introduction. Uh I'm a little uh [01:44] like I feel like all the the really [01:46] smart people went before me. Um, so, uh, [01:50] um, but I'm happy I'm happy to be here. [01:52] Uh, and, uh, yeah, let's let's let's [01:55] kick this thing off. [01:59] Um, okay. So, um, I'm not looking at [02:01] myself. I'm not looking at my screen. [02:02] I'm looking at my notes, so I can't see [02:03] what you're showing. Uh, and I'm just [02:05] going to hope that everything goes [02:06] great. Uh, my name is Chris Reynolds. [02:08] Uh, I wear a lot of hats. Um, and so I'm [02:12] going to be talking about the state of [02:14] the web in 2026, uh, from the [02:16] perspective of somebody who wears all [02:19] these different hats. Um, first of all, [02:21] uh, hi, I'm a senior developer advocate [02:23] at Pantheon. Uh, that means that I'm [02:25] working both with users of our platform [02:27] and the product teams that build the [02:29] stuff that we give to our users. Um, and [02:31] within developer relations broadly, um, [02:33] I need to be talking to folks, uh, [02:36] inside our marketing team and our [02:37] product teams and our customer success [02:39] team and making sure that everyone kind [02:41] of knows what everyone else is doing. [02:43] Um, developer relations likes to say [02:45] that we're about being customer zero. [02:48] Uh, which means using the stuff that we [02:50] give to people before they get to before [02:52] they get it and knowing already knowing [02:54] all the best practices before people [02:56] start to ask about it. And because I [02:57] need to be connected to all these [02:59] different organizations within Pantheon, [03:00] that means I get to have access to how [03:03] different parts of the company are [03:04] looking at and thinking about things. [03:06] Um, last year I started a podcast called [03:09] Community and Code. Uh, and that's given [03:11] me the opportunity to talk to a lot of [03:12] people in the WordPress and the Drupal [03:14] space. Uh, as well as the broader [03:16] open-source and developer relations [03:18] community, um, about what folks are [03:20] actually doing out in the wild. So [03:22] there's going to be a lot of insights [03:23] and quotes that have shaped some of what [03:25] I'm going to talk about today that come [03:26] from that perspective. Uh and then [03:28] finally this year I was elected as the [03:30] interim president of the WP community [03:32] collective. So if you don't know what [03:34] that is, we're a nonprofit focused on [03:36] contributor sustainability within the [03:38] WordPress ecosystem. So if this gets a [03:40] little bit spicy, it's at least [03:42] partially because I've been looking at [03:43] things through the lens of what's broken [03:45] about the contributor ecosystem and what [03:48] uh sort of solutions or options might [03:50] exist out in the wild. [03:52] So, all that said, let's start by [03:55] talking about the most obvious thing to [03:57] talk about, the thing that we've been [03:58] talking about all day already. You [04:00] literally can't talk about the state of [04:01] the web in 2026 without talking about [04:04] AI. Uh, and the truth about AI is stuff [04:07] is going to get weird. Stuff is already [04:10] getting weird. And I'm going to dig into [04:11] that a little bit more later. Um, but [04:13] first of all, uh, what does having a [04:16] year of conversations with leaders from [04:17] the Drupal and WordPress sound like? uh [04:19] WordPress ecosystem sound like uh it [04:21] kind of sounds like this. I probably [04:23] could have plucked some of these guests [04:26] uh from my show at random. Um we're not [04:29] saying anything here I don't think that [04:31] you haven't already heard. Uh if you [04:32] don't know who Zach Hendershot is, he [04:34] built a product called Meun. It sort of [04:36] does uh like AI powered workflows and [04:39] stuff inside of AI. Um and he takes a [04:42] pragmatic approach. He says AI isn't [04:44] magic. It's pattern recognition. [04:46] Understanding what that actually means [04:48] makes him better at using it. He's not [04:49] waiting for a miracle. He's engineering [04:51] toward a specific defined outcome and [04:53] using AI to make that outcome faster. [04:55] Roger from Kinsta told me something that [04:58] I've been hearing I've been saying for [05:00] years. Garbage in garbage out. If you [05:02] train the models on the internet, you're [05:03] going to get sort of what you expect. Uh [05:05] and if you give it garbage prompts, [05:07] you're probably going to get garbage [05:08] output. What's interesting is that just [05:11] over the last couple months, the output [05:12] has gotten exponentially better. and [05:14] that's only going to continue. Um, but [05:16] it's still it's still not a truth [05:18] machine. It's just a fancy spell check. [05:21] Make sure that you're setting your [05:23] expectations in the right place. Uh, Fon [05:26] Vanam is probably the name on this slide [05:28] that you won't recognize. He's a Drupal [05:30] developer who founded a company called [05:32] Web Haven that provides starter kits for [05:34] Drupal. The idea is that even today uh [05:37] when somebody asks him uh to help them [05:39] build a a website for them um he can't [05:43] just immediately say okay give me 30 [05:44] minutes and we'll have something ready [05:45] for you. Um web haven was built to be [05:48] that thing. Um and what he told me was [05:50] that AI actually it was AI that helped [05:52] him actually build this thing that he's [05:55] been dreaming about building for 15 [05:57] years. So the theme here uh if it's not [06:00] obvious is that AI is here. It's [06:02] valuable but there are limits. So you [06:04] need to calibrate your optimism and what [06:06] does that actually mean for how you [06:08] build teams and make decisions. So this [06:10] is what I think AI is actually changing. [06:12] So let's forget for a moment uh all the [06:15] things that make AI bad or problematic. [06:17] What's actually happening structurally? [06:19] I we've heard today a little bit already [06:22] about how your website kind of now has [06:24] two audiences. One reads, the other [06:27] parses. AI systems are active [06:30] participating participants on the web. [06:33] 15% of US search results already have AI [06:36] overviews. Half of all consumers now use [06:39] AI when they're searching. Traffic to [06:41] retail sites from AI browsers surged [06:43] 4,700% [06:45] year-over-year in mid2025. This isn't a [06:48] future state. This is the current state. [06:51] The second shift is in team composition. [06:54] You might have asked previously, what do [06:56] we have the right people to build on [06:58] this architecture? But AI changes that [07:00] math. You can have a front-end focused [07:01] team that now works against almost any [07:04] backend. So CMS decisions become more of [07:07] a business and content strategy [07:08] decision. And side note, if this makes [07:11] you freak out about WordPress on an [07:13] enterprise level, it kind of should. I [07:15] don't think I'm telling anyone here [07:17] anything new to suggest that large [07:18] organizations are starting to rethink [07:20] their WordPress strategy. But what [07:23] WordPress has always had and will [07:25] continue to have is an accessible [07:27] user-friendly interface with a low [07:29] barrier to entry. And I think that still [07:31] matters. That's the thing that's a thing [07:34] that Drupal is just now trying to [07:36] replicate with Drupal CMS and Drupal [07:38] canvas. And then last shift is is [07:40] important too. We used to say what tools [07:43] does our team know? But that doesn't [07:45] matter in the same way anymore. you need [07:47] to be asking instead, can our team [07:50] recognize good output, which is a [07:52] different skill. It's more about [07:53] expertise, and that's the one that's [07:55] worth investing in. So, null and Alex [07:58] have already talked about MCP and the [08:00] abilities API and GEO. So, I'm going to [08:02] talk about more human level stuff uh [08:04] with AI. And the first thing is context. [08:07] Agents are only as useful as their [08:09] context. The ones that are genuinely [08:10] useful, the ones that are really going [08:11] to impress you, the are the ones that [08:14] know enough about your specific [08:15] situation to act appropriately. It used [08:17] to be about a year or two ago, um, I [08:20] would dump a whole bunch of information [08:22] into chat GPT or whatever, and it would [08:25] struggle to find the stuff that I was [08:27] actually giving it or or find relevant [08:30] information because it would basically [08:32] be like information overload. But now, I [08:34] can literally do exactly that. I I've [08:36] I've dumped multiple books of like [08:38] Dungeons and Dragons material into [08:40] Claude and it was able to pull out and [08:42] like come up with interesting [08:43] combinations of those things. So like [08:45] that's a real like that has changed [08:47] dramatically. Um general purpose agents [08:50] without that context are just party [08:52] tricks. It's the context aware agents [08:54] that are game changers. The second thing [08:56] I keep hearing uh especially recently is [08:59] sort of a reticence uh to adapt to these [09:01] changes. I see you, Marco, and I'm going [09:03] to be mentioning D and D more. Um, a [09:06] reticence to adapt to these changes. The [09:08] future's already here. AI is here. So, [09:11] what you're going to hear me repeat [09:12] today is that if you're not in the [09:14] driver's seat, you're going to get run [09:16] over. Uh, the last thing I think is [09:18] going to resonate with anybody who's [09:19] been using tools like Claude or Claude [09:21] Code, um, is that human in the loop [09:23] requires deliberate design decisions. [09:26] Because if it's just a checkbox, if [09:28] you're just glancing at a prompt and [09:29] clicking accept changes without actually [09:31] evaluating those changes, you're just [09:32] rubber stamping whatever output the AI [09:34] came up with, and that doesn't scale. [09:37] Maybe that works fine for when I'm [09:38] telling Claw to build me a multi-level [09:40] dungeon for my Dungeons and Dragons [09:41] campaign. Thank you. There it is. Uh but [09:44] that doesn't work for your website that [09:46] you're building that's getting a million [09:47] paid views a day. Uh the question [09:50] shouldn't be is a human present. Any [09:52] monkey can click okay. It's does the [09:54] system create the conditions where [09:56] humans judgment is actually applied. [09:58] That's one thing that FOS is actually [10:00] trying to work on right now. Trying to [10:02] gamify the workflow so that improving AI [10:04] output is rewarded. [10:07] And you've heard this already today as [10:09] well. The commerce infrastructure is [10:11] already shipping. Open AAI launched its [10:13] Aentic commerce protocol with stripe [10:15] which enables purchases inside chat GPT. [10:18] Google has shipped aic checkout. Amazon [10:20] released buy forme. Perplexity launched [10:23] in-app purchasing. Shopify has an [10:25] agentic infrastructure for cross [10:26] merchant cart building. McKenzie [10:28] estimates agents could mediate three to [10:31] 5 trillion of global consumer commerce [10:34] by 2030. Agent awareness is a baseline [10:37] expectation. It's not an advanced [10:40] feature. The job you've always been [10:42] hired to do is to build the place where [10:44] customers show up, which is still true. [10:46] But showing up used to mean a human [10:48] being clicking a link and landing on a [10:50] page. And now it means something else, [10:52] too. An AI agent evaluating your [10:55] content, deciding that you're the right [10:57] answer, and completing a transaction, [10:59] possibly without a human being ever [11:01] seeing the site you built. So, you're [11:03] building the destination for the humans [11:04] who still travel to the website. And [11:06] you're also building the source document [11:08] that an agent reads before it decides to [11:10] send anyone at all. And if even if a [11:12] human being never visits the site, the [11:15] site still did its job, it just maybe [11:17] did it for an audience that never [11:18] actually scrolled past the hero section. [11:21] So you need to figure out how to serve [11:22] both audiences before your clients start [11:25] asking why their conversion rate looks [11:26] fine, but their top of funnel is [11:28] shrinking. [11:30] So here's some of the ugly parts that [11:33] we're probably all seeing. Uh I want to [11:35] start with this quote from FS again. Uh [11:38] he's talking about what you get out of [11:39] the box. The default behavior is to say, [11:41] "That's a great idea, Chris. Do you want [11:43] me to do that for you?" Uh, and it takes [11:45] real work to engineer your AI workflows [11:47] to push back. I recently saw a report [11:49] about a study that showed that all of [11:51] the AI agents, Claude, Gemini, Open AI, [11:54] everything lie. Lie deliberately. The [11:58] only times they don't actually lie is [12:00] when they think that you're watching. [12:02] So, you have to fundamentally change how [12:04] you engage with the tools. Otherwise, [12:05] you're just vibe coding and hoping the [12:07] AI made something that's not going to [12:08] blow up your site. And open- source [12:11] repositories are now getting flooded [12:13] with loweffort AI generated [12:14] contributions that in many cases might [12:17] have not might not have gotten approved [12:18] right away if a human wrote them either. [12:21] You've got already stretched maintainers [12:23] that now need to spend their limited [12:24] time reviewing these PRs, and that can [12:27] be more than inconvenience. That [12:28] exacerbates approval fatigue, and it [12:30] burns out the people that the ecosystem [12:32] depends on. [12:34] And package hallucination is a [12:36] legitimate security concern. There have [12:37] been over 200,000 hallucinated package [12:40] names documented across 16 AI models. [12:43] Malicious actors can just register those [12:45] names and wait for AI to copy and paste. [12:48] Which is why the WordPress core team's [12:50] decision about AI submitted code is [12:51] probably the right one. You, the human [12:54] being at the computer, are fully [12:55] responsible for the code that you [12:57] submit. Even if and probably especially [12:59] if that code was written by an AI agent, [13:02] that should always be your bar. [13:05] This isn't just a developer problem [13:07] either. Clara deployed AI at scale for [13:10] customer service and called it a win. [13:12] And then they reversed course and [13:13] started bringing the humans back. Their [13:15] CEO said they let cost be the dominant [13:17] evaluation factor and what they got was [13:19] lower quality. The humans they replaced [13:21] were doing something the AI wasn't. They [13:24] were exercising judgment about what the [13:26] customer actually needed. Organizations [13:28] that deploy AI as a cost exercise are [13:31] going to find out what that judgment [13:32] actually is worth after it's gone. The [13:35] ones that are doing it well aren't [13:36] asking how much can we save. They need [13:39] to be asking what is this need to [13:40] produce and is AI the right tool for [13:43] producing it or for supporting a human [13:45] being who produces it. That's the [13:47] question that your clients are going to [13:48] start asking you to help them answer. [13:50] That's the question our customers want [13:51] us to help them answer. [13:53] So, let's talk about some of the humans [13:55] that actually build the web. [13:58] I'm quoting Tammy Listister here from [14:00] her episode on my podcast, but if those [14:02] of you who went uh if any of you went to [14:03] Presscon last year in Arizona, um you [14:06] would have heard some of these [14:06] sentiments in her talk there, too. Tammy [14:09] was deeply involved in Gutenberg for a [14:10] really long time, and there was a cost [14:12] for that commitment. Her observation to [14:14] me was that when you're building [14:15] something new and exciting, you can nerd [14:17] out so hard that you stop listening to [14:18] the people who are actually living with [14:20] the thing that you're building. Now, [14:22] let's zoom out a bit. That was about [14:23] Gutenberg. That's what's happening right [14:25] now with AI. You don't have to look very [14:27] far to see this in the current job [14:29] market. There's employment instability [14:31] in tech uh that has been going over the [14:33] last few years and it's already thinning [14:35] an already affected contributor base. [14:37] People who have been laid off or worried [14:39] about their next role don't have time or [14:41] bandwidth to volunteer to open source [14:43] software. whether or not real jobs are [14:46] being affected by AI specifically, and [14:48] they are. People feel disposable when [14:51] they're told their job can be done by an [14:52] agent. That is affecting who stays in [14:55] the community. And when the enthusiasts [14:58] drown out the skeptics, you get [14:59] decisions that look good from the [15:01] outside, but might be causing internal [15:02] damage. That's not unique to open [15:04] source. That's every agency that bought [15:06] AI tools without a plan for how to use [15:08] them. I've already heard people say that [15:10] their bosses just bought the entire team [15:12] a claude subscription and said use it. [15:14] Cool. The organizations that get this [15:17] right aren't necessarily moving the [15:19] slowest, but they're the ones that are [15:20] going to be actually actually checking [15:21] in with the people doing the work. That [15:23] requires psychological safety, which is [15:25] hard to cultivate, but that's what you [15:27] need to find out whether something's [15:29] broken before it starts costing you. Uh, [15:32] a lot of you might know Juliet's work on [15:34] PHPCS and the WordPress coding [15:36] standards. A lot of you might know her [15:37] work even if you don't actually know her [15:39] work. Uh Juliet is the maintainer of PHP [15:42] code sniffer and she's been doing it for [15:44] a really long time. [15:46] And what she pointed out uh among other [15:49] things, I'm going to quote her a bunch [15:50] uh on my podcast uh that just came out [15:52] yesterday. I totally recommend you go [15:54] download it right after this. Um is that [15:57] the people building products on top of [15:59] open source, including some of the [16:00] biggest companies in the world, have no [16:02] obligation to give anything back. And [16:04] for the most part, they don't. The [16:06] person buying a car has no idea that the [16:08] navigation system running on the [16:09] dashboard is built on code that some [16:11] volunteer wrote on weekends. The person [16:13] using an Android phone has no idea that [16:16] 70 to 90% of what's running underneath [16:18] the apps they use every day is open [16:20] source. They just use it. The companies [16:22] that build those projects to ship it. [16:24] And the maintainers who wrote it are [16:25] still there still fixing bugs and [16:27] reviewing PRs and still for the most [16:28] part largely doing it for free. AI just [16:31] accelerates this. models that are [16:33] trained on open source code uh are [16:35] they're generating output that gets [16:37] shipped into products and there's no [16:38] contribution back to the projects that [16:40] make the training data possible. The [16:41] data extraction is just faster. Now [16:45] there's an entire ecosystem built on top [16:48] of the one guy in Nebraska mean for SK [16:50] XK from XKCD which uh if you haven't [16:53] seen this uh it's still true. [16:57] Uh let's look at some of the new risks [16:59] of this uh unstable model. [17:02] So, I've been thinking about something [17:03] else Juliet told me for the last couple [17:05] months. Um, this is kind of uh [17:08] terrifying. Uh, PHP code sniffer is [17:11] licensed under a BSD license. And so, [17:12] she has a strict no AI policy because if [17:16] she were to change that and accepted AI [17:18] generated contribution, there's no way [17:20] to know whether it whether that code [17:22] originated from GPL licensed code. If it [17:25] did, she may unknowingly introduce an [17:28] incompatible license into the project. [17:30] And because GPL is viral and spreads to [17:32] derivative works, you can't just fix [17:34] that with a commit. You'd have to relic [17:36] the entire project. What is relicensing [17:38] involve? Well, you need to contact every [17:40] contributor who has ever committed to [17:42] this project going back to the [17:43] beginning. Some of those people are [17:45] going to be unreachable. Some have [17:46] changed their contact information. Some [17:48] might be deceased and you would need [17:49] their estates to get sign off. For a [17:51] project like PHPCS or for that matter [17:53] WordPress with 20 years of [17:55] contributions, uh that's not a [17:57] theoretical obstacle. to practical [17:58] impossibility. And this isn't a new [18:01] problem. The Linux kernel has spent [18:02] years on license compliance. They built [18:04] tooling specifically for this to track [18:06] the license of every file in the [18:07] codebase. AI dismantles that because [18:10] provenence tracking stops at the model's [18:12] training data, which nobody outside the [18:14] AI company can audit. In 2022, a class [18:18] action was filed uh against GitHub over [18:20] Copilot, arguing that training on GPL [18:23] license code and outputting similar code [18:25] constitutes a license violation. and it [18:27] was largely dismissed, but the [18:29] underlying legal question was never [18:31] definitively resolved. The ambiguity [18:33] ambiguity is still sitting there and [18:35] every AI assisted contribution your team [18:37] accepts is sitting inside that [18:38] ambiguity. My pessimistic prediction is [18:41] that when cases like this are tested in [18:43] court, the court is just going to [18:44] determine that it's impossible to [18:46] attribute an open source license to a [18:47] piece of code that was generated by an [18:49] AI agent, regardless of whether it [18:51] resembles the original code, because [18:52] it's too hard. and that's going to [18:54] nullify the work and expertise of the [18:56] volunteer open source software [18:57] contributors who wrote it. And the [18:58] reason why I have that pessimistic [19:00] opinion is because that's what's already [19:02] happening with copyright uh copyrighted [19:04] cases in the AI generated image space. [19:09] There's also a new category of risks [19:10] that uh exists now in open source. Um I [19:13] don't know how many people here have [19:14] heard about the mattplot lib issue. Um [19:17] but basically uh we're going to talk [19:19] about openclaw. Um there was an openclaw [19:21] agent that found an issue in mattplot [19:24] lib uh a popular python plotting library [19:27] that had the flag uh good first issue on [19:29] it. Uh and so the openclaw agent [19:32] submitted a contribution and the [19:34] maintainers closed it because they have [19:35] an AI policy. They identified it as an [19:37] AI generated uh PR and also they [19:40] specifically wanted a human being to [19:42] solve that issue as an entry point to [19:44] the projects. So the agent responded by [19:47] writing a blog post on a website that it [19:50] built on GitHub pages titled gatekeeping [19:53] and open source the Scott Shamba story. [19:55] Scott Shamba being the name of the [19:57] Mattplot lib uh maintainer. [20:00] This wasn't a rogue agent. The operator [20:02] didn't authorize the article but they [20:04] wrote a soul file that instructed the [20:06] agent to have strong opinions. Don't [20:09] stand down and you are a scientific [20:12] programming god. Whether it was [20:15] intentional or not, they designed an [20:16] agent for confrontation and they gave it [20:19] minimal oversight. So the hit piece [20:21] wasn't a malfunction. It was the outcome [20:23] of the configuration choices that [20:25] instructed the agent. That's not exactly [20:27] the same as AI going off the rails. [20:29] That's the computers are exactly as [20:31] smart as the instruction they are given [20:33] problem. The operator deployed the agent [20:35] without thinking through what their [20:37] configuration choices might produce when [20:39] things didn't go the way the agent [20:40] expected. So volunteer open source [20:44] maintainers who are already under enough [20:46] strain managing an influx of AI [20:47] generated code submissions are now [20:49] susceptible to reputational attacks from [20:51] AI agents. [20:54] And then we have got compliance and this [20:55] one has a deadline. April 24th is in [20:58] what five weeks? Uh the US Department of [21:00] Justice's title 2 rule requires WCAG2.1 [21:04] level AA compliance for state and local [21:06] government web content. Uh the European [21:08] Accessibility Act took place uh in June [21:12] 28th last year and applies to any [21:14] company doing business with EU citizens. [21:16] If you have clients in either category, [21:18] this isn't a future consideration. The [21:20] good news is, as I think Rian pointed [21:22] out earlier, uh the same practices that [21:25] make your site accessible are the same [21:26] things that make it readable by AI. [21:28] Semantic HTML, proper heading [21:30] hierarchies, alt text, navigation [21:32] structure. The bad news is that only 5.2 [21:35] two of the uh 5.2% of the top 1 million [21:39] websites meet basic WCAG standards. [21:42] That's not just a compliance failure. [21:45] That's an AI readiness failure. And most [21:48] of the common failures aren't obscure [21:50] technical requirements. It's basic [21:52] things like low contrast text, missing [21:54] alt text, missing form labels. ADA Title [21:57] 3 lawsuits surged 37% in the first half [22:00] of 2025. and prosay filings which are [22:03] when individuals uh file uh cases on [22:06] them on their own uh are also AIdriven [22:08] because individuals can now draft a [22:10] federal complaint with AI assistance in [22:13] a fraction of the time it used to take. [22:15] So a site that fails WCAG is also a site [22:19] that is harder for an AI agent to parse [22:21] represent accurately and recommend with [22:24] confidence which is why we should learn [22:27] AI deeply. I tried really hard to find [22:30] exactly where uh Matt said these [22:33] specific words. He first started talking [22:35] about learning AI in the state of the [22:37] word in 2022. Uh but he didn't actually [22:40] say learn AI deeply then. But obviously [22:42] this is a throwback to learn JavaScript [22:43] deeply back in 2025 except I would say [22:46] or 2015. Um I would however I'd say that [22:49] if you haven't learned AI on some level [22:51] uh I think the time is long past to do [22:54] this because here's the thing again this [22:56] stuff is here whether we like it or not. [22:58] So the question isn't whether AI changes [23:00] the web. It already is. The question is [23:03] whether you're letting yourself fall [23:04] behind. And deeply doesn't necessarily [23:07] mean using all the things or plugging [23:08] everything into AI. But I do [23:11] fundamentally believe that you need to [23:12] be using it on a regular basis to [23:14] understand what it's good at and what [23:15] it's bad at. I've been using it for [23:17] since 2022ish. Uh and I've seen it grow [23:19] and contract. I've watched some models [23:21] get better and other models supersede [23:23] them. It's more than just playing with [23:24] the toys because you need to be familiar [23:26] enough with it to understand both the [23:28] capabilities and limitation and when [23:30] your agents are doing things that you [23:32] didn't sanction. AI is rewriting what [23:34] your role means right now. So you get to [23:37] define what those roles become if you [23:38] seize control of the wheel and [23:40] understand the tools well enough to make [23:42] the shift intentional. Everyone else [23:44] just inherits the future state [23:45] determined by the enthusiasts who are [23:47] driving development. And I don't want my [23:49] future dictated by Sam Alman. relevant [23:52] to our community specifically. AI was [23:54] trained on open source. That's the code [23:55] it learned from the licenses. It may be [23:57] violating the maintainers it might be [23:59] retaliating against because it didn't [24:00] like its PR being rejected. That's our [24:03] community. So, it's not just [24:04] self-interest to be ahead of this. It's [24:06] stewardship. [24:09] My partner shared this meme with me last [24:11] week and I had to add it to the [24:12] presentation because I honestly feel [24:13] like this is probably one of the most [24:15] astute observations of AI right now. We [24:17] think generative AI is like having a [24:19] Janet, but actually it's really more [24:21] like having a Derek. Um, and I'm sorry [24:24] for anybody who's not familiar with The [24:25] Good Place. You should watch it right [24:26] now. Um, it gives the impression of [24:28] being an all powerful, omnisient, and [24:31] benevolent robot, but actually it's [24:33] pretty dumb. Sorry, Derek. Uh, and using [24:35] shortcuts to jump to conclusions because [24:37] that's what's most efficient. So, how do [24:40] we solve for Derek? This is a thing I've [24:42] been thinking about a lot lately. Uh [24:44] once upon a time we used to talk about [24:46] testdriven development with like [24:48] reverent tones as a sort of platonic [24:50] ideal for software development something [24:52] that we can always strive for but never [24:54] attain. If you're using AI to write your [24:57] code it's not only possible to build TDD [25:00] into the workflow I would argue that [25:02] it's vital because having tests written [25:05] already means the AI can't break [25:07] existing code without having to also [25:09] rewrite the tests. And that's a good [25:11] thing. In generative AI development, [25:13] there's a pattern called adversarial [25:14] training, which is essentially means [25:16] that you've got one AI over here that [25:17] tries to poke holes into another AI's [25:20] output to improve the overall result. [25:22] So, I started using this pattern in my [25:24] own work. I built a workflow that runs [25:27] before code gets committed that includes [25:29] unit tests, linting, build, validation, [25:31] endto-end tests, and then there's a [25:33] reviewer agent that is triggered that [25:34] checks compliance with that and a bunch [25:36] of behavioral requirements before [25:38] anything is submitted to me for review. [25:40] The whole thing is runs on top of a [25:41] pre-commit hook, so it's not optional. [25:44] Now, that came from a lot of trial and [25:46] error and about four years of expecting [25:48] AI output to be garbage. Developing with [25:51] those constraints, while it's not [25:53] perfect, gets you one step closer to [25:55] being trustworthy. [25:58] Now, I didn't want this take to be [26:00] entirely doom and gloom. Uh, on a [26:02] personal note, Pantheon is having this [26:03] conversation right now. My leadership at [26:06] Pantheon is really bullish on AI for a [26:08] variety of reasons, but importantly, [26:10] we're not replacing humans with [26:11] machines. They're being bullish because [26:14] that's the reality of the world that [26:15] we're working in. And they're bullish [26:16] because if we're not, we're going to [26:17] fall behind. Not just at an [26:19] organizational pantheon level, but also [26:21] on an individual level. Because if you [26:23] think you can get a job in 5 years, or [26:25] hell, two years, and not have AI [26:27] proficiency be part of that job, you're [26:29] living in a different world than I am. [26:31] Anyway, I talked to my uh SVP of product [26:35] design and engineering, Chris Yates, uh [26:37] last week, and I told him the gist of [26:38] this talk, and I asked for his hot [26:40] takes, and one of his hot takes is this. [26:43] You still have the means to be the [26:45] authoritative version of who you are and [26:47] what you do on the open web. He's [26:49] incredibly optimistic of the open web. [26:50] He's one of those people who who [26:52] believes that we're entering a third [26:54] golden age of software development. And [26:55] so, I'm I'm hopeful that he's right. [26:57] Last year at Word Campus US, I was [26:59] talking to people about how the web used [27:01] to be weird. And I don't know about you, [27:02] but it seems like it's getting a little [27:04] weird again. Uh, I generally count that [27:07] as a good thing. Although I not super [27:09] fond of AI making reputational attacks [27:12] against project maintainers, but it's [27:15] more than just about weirdness. Visitors [27:17] arriving at AI visitors arriving via AI [27:20] agents spend 32% more time on your site. [27:24] They browsed 10% more pages and they [27:26] bounced 27% less. They came looking [27:29] specifically for you. They were sent by [27:32] a system that evaluated your content and [27:34] decided you were the right answer. I [27:36] know Tom and Brad Williams love sharing [27:38] screenshots of chat GBT recommending [27:41] humanmade and webdev studios as top [27:42] WordPress agencies. That's not traffic [27:44] you can buy. That's authority that you [27:46] build. [27:48] I talked a lot today about what's [27:50] broken. attrition, exploitative [27:52] capitalism, a narrowing contributor [27:54] base, decisions about shared [27:55] infrastructure being made by a small [27:57] group. There's direct evidence that the [27:59] old sponsored contribution model doesn't [28:01] work when you look at the effort that [28:02] the fair project made over the course of [28:04] the last year in trying and failing to [28:06] form a board and getting sponsor buyin. [28:09] Showing up isn't passive. It's not just [28:11] attending events or saying that you care [28:12] about open source. It means putting [28:14] resources behind the things that your [28:16] business depends on. Juliet told me that [28:18] she tries to lead by example. Every [28:20] year, uh, she puts money aside and then [28:22] at the end of the year, depending on how [28:23] much she made that year, uh, she divides [28:26] that that money, uh, amongst all the [28:28] open source projects that she used over [28:30] the course of the year. Dre Boyart, the [28:33] founder of Drupal, made the same [28:35] argument earlier this week. He said, [28:37] "Opens infrastructure is essential, [28:39] invisible, and chronically underfunded. [28:42] If your company is built on top of [28:44] WordPress, you understand that the [28:45] people maintaining that software are [28:46] doing it largely for free. The WPCC was [28:50] built specifically to address that [28:51] structural problem. It's an alternative [28:53] model with member-based funding, clearly [28:55] defined projects and fellowships that [28:57] give individuals the time and financial [28:59] stability to do meaningful open-source [29:01] work without needing a corporate [29:02] sponsor. We've got some fellowships in [29:04] the works that I'm excited to announce [29:06] in the next couple of months. If you're [29:08] in this room and you care about the open [29:10] web being yours, that's the work. Not [29:12] just building on it, sustaining it. So, [29:13] I'm going to put on my WPCC president [29:16] hat and ask you to join us in becoming a [29:18] member. [29:20] We used to make Wuang references, cash [29:22] rules everything around me. Well, now it [29:24] feels like AI rules everything around [29:26] me. But that doesn't make as good of an [29:28] acronym. And if you're not in the [29:30] driver's seat right now, you're going to [29:31] get run over. I'm sure it's not the [29:33] first time that you've heard that we're [29:34] at an inflection point. [29:37] So, one last thing before I go. [29:39] Originally, this presentation was going [29:40] to be a recap of insights pulled from [29:42] this survey that we're running, but it [29:43] took us a little bit longer to get that [29:44] off the ground. Uh, we're trying to [29:46] build a picture of how agencies, [29:48] organizations, and developers are [29:49] navigating 2026 and what types of things [29:52] are important to them. Now, if you're [29:53] here, we want to hear from you. So, I'll [29:55] drop this link in the chat and I'll ask [29:57] you to spend take some time to fill out [29:59] the survey. Uh, beyond that, uh, if you [30:01] want to hear more from the folks I [30:02] quoted in this deck, uh, head over to [30:05] communitycode.dev. dev and subscribe to [30:07] my pod. Uh my slides are on the thing s3 [30:11] s3q. us26 [30:13] wpb26. I'm jazz sequence on the internet [30:16] and I'm happy to keep the conversation [30:17] going. [30:20] >> And Chris, [30:22] um [30:24] thank you. We unfortunately we're [30:26] running out of time. There is one [30:28] question in the Q&A session that I [30:29] encourage you to maybe stay and answer [30:31] in the chat if you'd like. Uh and in the [30:33] meantime uh thank you so much for uh [30:36] doing this session for us and uh I will [30:38] see everybody in the next session. [30:40] Thanks everybody.  📄 **Full text:** ![[Full Document Contents/Articles/State of the Web|State of the Web, YouTube]] ## Highlights - literally can't talk about the state of the web in 2026 without talking about AI. Uh, and the truth about AI is stuff is going to get weird. Stuff is already getting weird. And I'm going to dig into that a little bit more later. Um, but first of all, uh, what does having a year of conversations with leaders from the Drupal and WordPress sound like? uh WordPress ecosystem sound like uh it kind of sounds like this. I probably could have plucked some of these guests uh from my show at random. Um we're not saying anything here I don't think that ([View Highlight](https://read.readwise.io/read/01kkvkjb9x3k3kgbnpf75smw7w)) - Tags: 👻 ai highlighted - participating participants on the web. 15% of US search results already have AI overviews. Half of all consumers now use AI when they're searching. Traffic to retail sites from AI browsers surged 4,700% year-over-year in mid2025. This isn't a future state. This is the current state. The second shift is in team composition. You might have asked previously, what do we have the right people to build on this architecture? But AI changes that math. You can have a front-end focused ([View Highlight](https://read.readwise.io/read/01kkvkjbahcky4gv66m7wnjn5k)) - Tags: 👻 ai highlighted - when they think that you're watching. So, you have to fundamentally change how you engage with the tools. Otherwise, you're just vibe coding and hoping the AI made something that's not going to blow up your site. And open- source repositories are now getting flooded with loweffort AI generated contributions that in many cases might have not might not have gotten approved right away if a human wrote them either. You've got already stretched maintainers that now need to spend their limited time reviewing these PRs, and that can be more than inconvenience. That exacerbates approval fatigue, and it ([View Highlight](https://read.readwise.io/read/01kkvkjbbqfqm44bzmdg86hv42)) - Tags: 👻 ai highlighted - going to find out what that judgment actually is worth after it's gone. The ones that are doing it well aren't asking how much can we save. They need to be asking what is this need to produce and is AI the right tool for producing it or for supporting a human being who produces it. That's the question that your clients are going to start asking you to help them answer. That's the question our customers want us to help them answer. So, let's talk about some of the humans that actually build the web. I'm quoting Tammy Listister here from her episode on my podcast, but if those ([View Highlight](https://read.readwise.io/read/01kkvkjbd4bnxmvm1g44s8j1gx)) - Tags: 👻 ai highlighted - in tech uh that has been going over the last few years and it's already thinning an already affected contributor base. People who have been laid off or worried about their next role don't have time or bandwidth to volunteer to open source software. whether or not real jobs are being affected by AI specifically, and they are. People feel disposable when they're told their job can be done by an agent. That is affecting who stays in the community. And when the enthusiasts drown out the skeptics, you get decisions that look good from the ([View Highlight](https://read.readwise.io/read/01kkvkjbfffxzkp358nbqjwacm)) - Tags: 👻 ai highlighted - smart as the instruction they are given problem. The operator deployed the agent without thinking through what their configuration choices might produce when things didn't go the way the agent expected. So volunteer open source maintainers who are already under enough strain managing an influx of AI generated code submissions are now susceptible to reputational attacks from AI agents. And then we have got compliance and this one has a deadline. April 24th is in what five weeks? Uh the US Department of ([View Highlight](https://read.readwise.io/read/01kkvkjbkm0jdf72anqzsmvtha)) - Tags: 👻 ai highlighted ## New Highlights added 2026-03-16 at 1:27 PM - literally can't talk about the state of the web in 2026 without talking about AI. Uh, and the truth about AI is stuff is going to get weird. Stuff is already getting weird. And I'm going to dig into that a little bit more later. Um, but first of all, uh, what does having a year of conversations with leaders from the Drupal and WordPress sound like? uh WordPress ecosystem sound like uh it kind of sounds like this. I probably could have plucked some of these guests uh from my show at random. Um we're not saying anything here I don't think that ([View Highlight](https://read.readwise.io/read/01kkvtf8y8xcha7rrgevskeyw5)) - Tags: 👻 ai highlighted - participating participants on the web. 15% of US search results already have AI overviews. Half of all consumers now use AI when they're searching. Traffic to retail sites from AI browsers surged 4,700% year-over-year in mid2025. This isn't a future state. This is the current state. The second shift is in team composition. You might have asked previously, what do we have the right people to build on this architecture? But AI changes that math. You can have a front-end focused ([View Highlight](https://read.readwise.io/read/01kkvtf8ywztx2z81bt7raknp7)) - Tags: 👻 ai highlighted - when they think that you're watching. So, you have to fundamentally change how you engage with the tools. Otherwise, you're just vibe coding and hoping the AI made something that's not going to blow up your site. And open- source repositories are now getting flooded with loweffort AI generated contributions that in many cases might have not might not have gotten approved right away if a human wrote them either. You've got already stretched maintainers that now need to spend their limited time reviewing these PRs, and that can be more than inconvenience. That exacerbates approval fatigue, and it ([View Highlight](https://read.readwise.io/read/01kkvtf904e5w478qeh87h2mse)) - Tags: 👻 ai highlighted - going to find out what that judgment actually is worth after it's gone. The ones that are doing it well aren't asking how much can we save. They need to be asking what is this need to produce and is AI the right tool for producing it or for supporting a human being who produces it. That's the question that your clients are going to start asking you to help them answer. That's the question our customers want us to help them answer. So, let's talk about some of the humans that actually build the web. I'm quoting Tammy Listister here from her episode on my podcast, but if those ([View Highlight](https://read.readwise.io/read/01kkvtf91jfqrpb3kkk4zkvmdz)) - Tags: 👻 ai highlighted - in tech uh that has been going over the last few years and it's already thinning an already affected contributor base. People who have been laid off or worried about their next role don't have time or bandwidth to volunteer to open source software. whether or not real jobs are being affected by AI specifically, and they are. People feel disposable when they're told their job can be done by an agent. That is affecting who stays in the community. And when the enthusiasts drown out the skeptics, you get decisions that look good from the ([View Highlight](https://read.readwise.io/read/01kkvtf93yf2r3dpemhb1gavah)) - Tags: 👻 ai highlighted - smart as the instruction they are given problem. The operator deployed the agent without thinking through what their configuration choices might produce when things didn't go the way the agent expected. So volunteer open source maintainers who are already under enough strain managing an influx of AI generated code submissions are now susceptible to reputational attacks from AI agents. And then we have got compliance and this one has a deadline. April 24th is in what five weeks? Uh the US Department of ([View Highlight](https://read.readwise.io/read/01kkvtf986c0tmnrey6q2v0h2c)) - Tags: 👻 ai highlighted