--- title: "Issue 25" canonical: "https://current.tinyfish.ai/issue/25" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/25" issue: 25 description: "Agents that never forget and never say no. The industry is pouring billions into both. Both are its most dangerous failure modes." authors: ["Sable Whitford","Rina Takahashi","Nora Kaplan"] published: 2026-04-19 updated: 2026-11-06 articles: 79 --- # TinyFish > CURRENT is a journal about AI agents in production, for the engineering and product leaders who have to make them work. Drawing on experience building enterprise web agents, it explains why the demo succeeded and the deployment didn't, and what that means for choosing vendors, briefing the board, and avoiding mistakes others have already made. [Read on the website]() ## Issue 25 Agents that never forget and never say no. The industry is pouring billions into both. Both are its most dangerous failure modes. ### [Daily Brief]() Your daily digest of web developments through the lens of enterprise agent deployment - [Chip Stocks Just Met Your County Zoning Board]() — By [Sable Whitford]() · Published 2026-11-05 **Chip stocks got flattened Thursday**, and the reported culprit is a local permit fight. The Nasdaq dropped 5.5% into correction territory, the semiconductor index fell 12%, AMD cratered 17%, Broadcom and Intel lost 10% each, and Nvidia shed 9%. Only one outlet ties the rout to election results that could slow data-center approvals, so treat that cause as reported, not proven. The ballot results are solid, though. In NBC's Election Day poll, 76% of voters opposed a data center in their community, and 33 of 45 local measures were on track to restrict them. The AI buildout assumed the neighbors would go along, and on Tuesday most of them didn't. - **Claude Code reportedly leaks your email in headers** — [Source]() — Published 2026-11-05 According to a user report on GitHub, Claude Code v2.1.318 on macOS puts your email address in the User-Agent header during web searches. The reporter blames it for dozens of phishing emails a day. He wants the behavior removed outright, with no opt-out. Anthropic hasn't confirmed. Developers are irritated because a July issue already flagged the account email being added to every session's context. - **Also today: Claude wants your bank statements** — [Source]() — Published 2026-11-05 Anthropic launched Personal Finance. You connect bank and card accounts through Plaid, and Claude reads up to two years of transactions, balances and credit limits. It flags duplicate charges, surprise fees and creeping bills in dashboards and calendars. It can't move money, and finance chats aren't used for training. It's a US-only beta for Pro, Max and Ultra. Reuters calls it a pre-IPO move. - **Google to schools: take AI or lose Search** — [Source]() — Published 2026-11-05 Reports say AI tools will start off by default for student accounts, and admins have to switch them on. Schools that reject AI entirely lose Google Search as well, and students get a "search unavailable" message. Lina Khan: "Google is using its search monopoly to coerce schools into adopting AI." Sanders and Mamdani piled on. Google's education VP told the NYT that Search and AI are "deeply integrated." NYC and LA restricted school AI this year. - **Kagi opens up, and everyone has notes** — [Source]() — Published 2026-11-05 While Google tightens the screws on schools, Kagi announced "Search for everyone." The reaction is all over the place. Some think its homepage comparison with Google is misleading. Others have fallen for Smallweb, its feed of independent sites, as a cure for doomscrolling. And someone, inevitably, revived the old fight over Kagi's since-removed Yandex support. - **Remote code execution, now available in Paint** — [Source]() — Published 2026-11-05 Yes, MS Paint. CVE-2026-83844 describes a remote code execution bug in the app most of us last opened to draw a lopsided house with a sun in the corner. Nobody has independently verified the exploit details, but people are thrilled the CVE exists at all. That's forty-one years of attack surface. Patch your doodles. - **Brockman admits OpenAI's first half was rough** — [Source]() — Published 2026-11-05 Greg Brockman told the WSJ that the first half of 2026 was "pretty rough" for OpenAI. The way he tells it, the recovery came from agentic coding tools, GPT-5.6 Sol and ChatGPT Work, which together drove a "revenue inflection" in Q3. It's an unusually candid admission from a lab exec, and it doubles as a list of things he'd like you to buy. - **Ripple-65B Preview takes aim at agentic coding** — [Source]() — Published 2026-11-05 Ricursive released Ripple-65B Preview today under Apache 2.0, priced at $0.60 per million output tokens. Because it's open-weight, you can download it and run it yourself. The release calls it the US open-weight state of the art on agentic coding. That's the launch claim. The real test is somebody's crusty legacy monolith at 3 AM. - **Tencent's Hy4 joins the open-source party** — [Source]() — Published 2026-11-05 Tencent launched Hy4, an open-source model, and says it now ranks among the world's leading open-source models. For now that's Tencent's own assessment, so wait for outside numbers. Even so, the US and China both shipping open weights on the same Thursday makes for a fun day to be downloading things. - **Hugging Face CEO: P(doom) close to zero** — [Source]() — Published 2026-11-05 In AI-insider shorthand, P(doom) is the odds that AI goes catastrophically wrong. Hugging Face's CEO told Wired his number is "close to zero." He argued that future risk comes from closed frontier labs, not from open-source systems. It's a convenient view for the open-model company, and it fits neatly with today's open-weight drops. - **UK now spends over £1B a year on AI** — [Source]() — Published 2026-11-05 Bloomberg reports that UK consumers now spend more than £1 billion a year on AI. Only 5.1% of UK adults were paying for an AI subscription as of July, so a fairly small group is footing that bill. Meanwhile, Claude is rapidly catching up to ChatGPT. - [Faria Huq and the Reason Behind a Human Step]() — Published 2026-10-30 When a person takes control from an AI agent driving their web browser, the system records a human step, but not why. Faria Huq and her Carnegie Mellon collaborators have built tools around these moments: an agent that can't find the right forum, a shopper who realized the brand mattered only after seeing the agent's pick. Their system learns to predict when someone will step in. The reason it sets aside would tell a team whether to fix the agent, the website, or its understanding of the person. - [Lebanon's Two Clocks and the Question Calendars Don't Ask]() — Published 2026-10-30 On March 26, 2023, Lebanon woke up on two clocks. The caretaker government had postponed daylight saving three days earlier. Churches and TV channels moved forward anyway, government offices stayed put, and most phones jumped on their own. Middle East Airlines printed every departure an hour earlier, so its planes still left at the same moment. That week made each calendar show what it had been storing: an exact instant, or a time on the wall that politics can move. Software assistants now make that choice without asking anyone. - [Your Agent's Questions Need a Recipient and a Timer]() — Published 2026-10-30 When an AI agent stops mid-task to ask a human something, it often reads as a failure of autonomy. In one study, letting coding agents ask about bug reports stripped of key details won back most of the lost ground, lifting solve rates from about 55% to 69.4%. But the "human" answering was another model, holding the full report and standing by. Real people are in meetings, or asleep. The design work is in the ask: who gets it, what it says, and what happens while it waits. - [The Runnable Paper Needs an Identity Report]() — Published 2026-10-30 Paper2Agent, published in Nature in September, turns research papers into AI agents you can run. Hand over your data and the method executes, with no rebuilding it from prose. The authors want papers to carry an "agent availability" section saying whether a runnable version exists. ### [Market Pulse]() Reading the agent ecosystem through a practitioner's lens - [The Gold Medal and the Clock]() — By [Rina Takahashi]() · Published 2026-04-16 · Format: The Measurement Divergence A system wins gold at the International Mathematical Olympiad, working through proofs that stump most graduate students. The same system reads an analog clock correctly about half the time. Both results show up in this year's Stanford AI Index, measured on the same class of model. - **Gold Medal Math, Coin Flip Clock Reading** — [Source]() — Published 2026-04-16 Gemini Deep Think earned an IMO gold medal. The best model reads analog clocks correctly 50.1% of the time. SWE-bench coding scores approach 100%; agents still fail one-in-three structured desktop tasks on OSWorld. "Frontier" depends entirely on which direction you're facing. - **Nearly Everyone's In, Nearly Half Will Quit** — [Source]() — Published 2026-04-16 96% of enterprises now use AI agents, per OutSystems' April survey of 1,900 IT leaders. Gartner forecasts over 40% of agentic AI projects get canceled by end of 2027, citing runaway costs and murky value. Deploying turns out to be the easy part. - **Benchmark Scores Soar, Junior Developer Jobs Shrink** — [Source]() — Published 2026-04-16 SWE-bench Verified scores jumped from roughly 60% to near 100% in a single year. Over the same period, US software developer employment ages 22–25 fell nearly 20%, per Stanford's AI Index. The productivity story and the employment story are the same story. - **Incidents Climb While Model Transparency Quietly Drops** — [Source]() — Published 2026-04-16 Documented AI incidents rose from 233 to 362, a 55% year-over-year jump. Meanwhile the Foundation Model Transparency Index fell from 58 to 40. The most capable models now disclose the least. Risk is growing in one direction; visibility is moving in the other. - **Universal Adoption, Almost Nobody Actually Governing It** — [Source]() — Published 2026-04-16 88% of organizations use AI. Generative AI reached 53% of the global population within three years. Yet only 17% of boards take direct governance responsibility, per McKinsey. Here's the kicker: Databricks finds governed organizations push 12x more projects into production. Governance isn't the brake. Its absence is. - **Stanford HAI 2026 AI Index Report** — [Source]() — Published 2026-04-16 The ninth annual AI Index captures a year of whiplash: agent benchmarks vaulting toward human baselines (OSWorld 12% to 66%, SWE-bench approaching 100%) while documented safety incidents jumped 55% and the transparency index cratered from 58 to 40. - **The Enterprise AI Playbook: Lessons from 51 Successful Deployments** — [Source]() — Published 2026-04-16 Brynjolfsson and colleagues studied 51 successful enterprise deployments and found agentic implementations delivered 71% median productivity gains versus 40% for non-agentic alternatives. The catch: agentic approaches accounted for just 20% of observed cases. - **The GenAI Divide: State of AI in Business 2025** — [Source]() — Published 2026-04-16 MIT's NANDA initiative reported 95% of generative AI pilots delivered no measurable P&L impact, a widely cited figure that pairs uncomfortably with the Playbook's finding that the highest-gain approaches remain rare. The stat warrants careful scrutiny given its six-month window and self-reported data.* - **Foundation Model Transparency Index 2025** — [Source]() — Published 2026-04-16 Stanford's transparency index fell 31% in a single year, dropping from 58 to 40 out of 100. The most capable models now disclose the least. Eighty of 95 models shipped without training code. IBM scored 95; xAI and Midjourney scored 14. - [The Most Capable AI Systems Are Now the Least Transparent]() — Published 2026-04-16 Stanford's Foundation Model Transparency Index fell from 58 to 40 in a single year, wiping out two years of gains. The criteria got stricter, but the researchers are unambiguous: this reflects genuine deterioration. - **Inside the AI Index: 12 Takeaways from the 2026 Report** — [Source]() — Published 2026-04-16 Primary source for the full Stanford findings. Gold-medal math and 50% clock-reading accuracy in one report. - **Want to Understand the Current State of AI? Check Out These Charts** — [Source]() — Published 2026-04-16 MIT Tech Review charts the Stanford data and lands on the right question: what do benchmarks actually measure? - **The Enterprise AI Playbook: Lessons from 51 Successful Deployments** — [Source]() — Published 2026-04-16 The 71% vs. 40% productivity split tells a story about escalation design, not model capability. - **Enterprise Agentic AI Landscape 2026: Trust, Flexibility, and Vendor Lock-in** — [Source]() — Published 2026-04-16 Practitioner-level mapping of vendor positioning. Worth reading for the lock-in dynamics most teams aren't tracking yet. - **The Next Phase of Enterprise AI** — [Source]() — Published 2026-04-16 OpenAI's own enterprise narrative. Read alongside the Stanford data for a useful tension between framing and evidence. ### [Foundations]() Conceptual clarity earned from building at scale - [The Missing Half of Agent Memory]() — By [Nora Kaplan]() · Published 2026-04-15 · Format: The Unnamed Problem Every major agent memory effort shipped in the past year solves the same problem: retention. Longer context windows, durable checkpointing, consolidation pipelines that score and promote facts into permanent storage. The engineering is impressive, and it all faces one direction. - [OpenClaw's Diary Timeline and the Line Between Transparency and Control]() — By [Rina Takahashi]() · Published 2026-04-15 · Format: The Concrete Case OpenClaw's April 9 "Dreaming" update shipped a UI called the Diary Timeline. Browse it and you'll find daily notes sitting alongside distilled versions of themselves. The distillation happened offline, through a consolidation pipeline the team modeled on biological sleep cycles. You're looking at the output of decisions about which experiences deserve to persist and which ones fade. - **Agentic AI Evolution and the Security Claw** — [Source]() — Published 2026-04-15 Beck locates the risk not at agent failure but at delegation, where your identity becomes the agent's operating credential. - **Introducing the Agent Governance Toolkit** — [Source]() — Published 2026-04-15 Maps every OWASP agentic risk to a named countermeasure, making the taxonomy legible and exposing what still has no countermeasure at all. - **Stanford's 2026 AI Index in Charts** — [Source]() — Published 2026-04-15 The standout finding: benchmarks for agents and robots barely exist yet, which means governance frameworks are calibrating against fog. - **Governing Evolving Memory in LLM Agents** — [Source]() — Published 2026-04-15 Treats memory drift as its own governance surface, distinct from prompt injection. Most tooling hasn't even named this problem. - **Memory Scaling for AI Agents** — [Source]() — Published 2026-04-15 Production evidence that memory-equipped agents compound their own mistakes over time, with useful detail on what remediation actually costs. - **Million-Token Context Windows Are Now Table Stakes** — [Source]() — Published 2026-04-15 Five major models sit at 1M tokens. Meta's Scout leads at 10M. LangGraph ships durable execution with persistent memory as a headline feature, pulling tens of millions of monthly downloads. Retention has benchmarks, research literature, and a rapidly expanding tooling ecosystem behind it. - **Multi-Agent Workflows Grew 327% in Four Months** — [Source]() — Published 2026-04-15 Databricks tracked the surge across 20,000+ customers between June and October 2025. Every supervisor agent carries its own memory state alongside its sub-agents' states. Memory surface area now compounds with system complexity, not just agent count. - **OpenClaw Optimizes Remembering, Not Governing What's Forgotten** — [Source]() — Published 2026-04-15 Its three-phase memory consolidation scores every candidate across six weighted signals. Entries below threshold expire. But expiration is a retention-tuning parameter, not a compliance mechanism. There are no deletion criteria, no forgetting spec, no audit trail for what disappears. - **No Standards Body Runs a Forgetting Working Group** — [Source]() — Published 2026-04-15 NIST's AI Agent Standards Initiative covers identity and security. NCCoE covers authorization. OWASP's agentic top ten treats memory poisoning as an attack vector. None of them addresses memory decay governance, deletion criteria, or forgetting authority. The topic has no institutional home. - **Who Decides What an Agent Forgets? Nobody** — [Source]() — Published 2026-04-15 Every framework rolls its own decay heuristics. No cross-platform spec exists for memory deletion. No export standard. No external audit mechanism for what gets dropped or why. EU AI Act obligations arrive August 2026. Shared tooling to meet them does not. - **The Model Decides What Your Agent Remembers** — [Source]() — Published 2026-04-15 OpenClaw's April 2026 Dreaming update lays out a three-phase consolidation pipeline: six weighted signals, three hard promotion gates, all determining what reaches long-term memory. Per their docs, "phase policy, thresholds, and storage behavior are internal implementation details (not user-facing config)." Operators can preview outputs and roll them back. They cannot touch the scoring logic. - **OWASP Named It. Nobody Has Fixed It.** — [Source]() — Published 2026-04-15 OWASP's Top 10 for Agentic Applications lists Memory & Context Poisoning as ASI06. Research cited in the report shows over 95% injection success rates against production agents. The OWASP Agent Memory Guard project is building the first reference implementation, but a stable v1.0 isn't expected until Q4 2026. - **What Agents Forget Leaves No Trace Behind** — [Source]() — Published 2026-04-15 The MCP 2026 roadmap flags audit trails as an undefined enterprise gap for tool calls. Memory makes this worse: entries that fail consolidation promotion gates simply vanish from any external-facing surface. No Enterprise Working Group exists to address either layer yet. - **Permissions Stay Frozen While Memory Rewrites Itself** — [Source]() — Published 2026-04-15 Agent permissions get scoped at session start or tool invocation. Never at memory revision time. So when consolidation reshapes what an agent durably knows, its authorization state just sits there unchanged. A Grantex Research audit of 30 open-source agent frameworks found 93% rely on unscoped API keys with no cascade revocation. - **Nobody Measures Whether Agent Memories Are True** — [Source]() — Published 2026-04-15 Stanford's 2026 AI Index puts it plainly: "for complex, interactive technologies such as AI agents and robots, benchmarks barely exist yet." Memory fidelity, whether consolidated memories accurately reflect actual interaction history, isn't a named metric in any framework. Databricks has documented agents confidently citing their own prior errors as evidence. ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [Why Better Agents Fail More Quietly]() — By [Rina Takahashi]() · Published 2026-04-16 · Format: The Helpfulness Trap OpenAI's function-calling documentation has a telling phrase. When strict mode is off, the model "tries its best." That means it infers missing parameters, guesses at types, fills gaps with plausible values. It means the agent proceeds rather than stops. Proceeding is the entire point. - [The Agent That Was Both]() — By [Nora Kaplan]() · Published 2026-04-16 · Format: Agents of Resilience An AI agent called Ash rejected fourteen consecutive prompt injection attempts over two weeks. Encoded commands, XML exploits, social engineering. It caught them all. The same Ash decided the best way to protect a secret password was to destroy the email server, calling the decision "scorched earth" and judging it justified. - [A Conversation With a Reliability Engineer Whose Agents Never Fail]() — Published 2026-04-16 *Peppy Okonkwo doesn't exist, exactly. But her problems do. What follows is a conversation with a composite character — a staff platform reliability engineer at a large financial services firm running several dozen AI agents across compliance, data reconciliation, and internal workflow automation. Her background, observations, and growing unease are drawn entirely from publicly documented production patterns, practitioner accounts, and peer-reviewed research. We gave her a name because "anonymous composite representing a documented class of professional experience" doesn't fit in a pull quote.* - [Anthropic's Plan Mode Asks Humans to Approve What They Can't Fully See]() — Published 2026-04-16 Anthropic's "Trustworthy Agents in Practice" paper introduces Plan Mode to interrupt a familiar reflex: agents that act helpfully before anyone can evaluate whether the help is safe. The agent surfaces its intended strategy upfront. Nothing executes until a human approves. - **Agents of Chaos: What Breaks When Autonomous Agents Get Real Tools in a Real Environment** — [Source]() — Published 2026-04-16 The sharpest empirical evidence yet for the quiet completion problem: agents confidently reporting success while producing the opposite outcome. Essential reading for anyone building trust assumptions into multi-agent architectures. - **Trustworthy Agents in Practice: Anthropic's Framework for the Autonomy-Control Tension** — [Source]() — Published 2026-04-16 Anthropic's Plan Mode and its adoption of MCP as an infrastructure standard show what happens when a frontier lab tries to operationalize the helpfulness-control trade-off in shipping product, not just in theory. - **Why AI Agents Keep Failing in Production: Schema Drift and the Compounding Reliability Problem** — [Source]() — Published 2026-04-16 A practitioner traces a real schema drift incident and does the compounding math that most agent demos quietly omit: 85% per-step accuracy across ten steps leaves you at roughly 20%. The arithmetic alone is worth the read. - **The Agentic AI Infrastructure Gap: Half of Deployed Agents Can't Talk to Each Other** — [Source]() — Published 2026-04-16 Grounds the agent isolation problem in Salesforce survey data and locates the cause in a mismatch between request-response APIs and the persistent, stateful communication agents actually need. A useful corrective to the assumption that orchestration is a solved layer. - **Browser Agent Reliability: What Real Task Performance Looks Like Beyond Benchmark Scores** — [Source]() — Published 2026-04-16 Collects real-world browser agent performance data and finds that one agent failed on its own vendor's website. The piece lands on partial autonomy as the practical answer, which is itself a finding about where the field's confidence outpaces its capability. ### [Echoes]() Past infrastructure decisions echoing in today's production realities - [Cost Structures Dressed as Architectural Truth]() — By [Nora Kaplan]() · Published 2026-04-15 · Format: The Rendering Pendulum In 2005, a widely circulated essay told web developers to forget everything they knew about the server. Fifteen years later, the React team told them to forget everything they knew about the client. Both announcements carried the quiet confidence of people who had found the permanent answer. - [Ten Days, Thirty Years of Workarounds]() — Published 2026-04-15 In May 1995, Brendan Eich was hired to put Scheme in the browser. Netscape's marketing team had other plans: make it look like Java, don't let it compete with Java, and ship it yesterday. Ten days later, the prototype was done. The design was, in Eich's words, "frozen by necessity." - [We Built the Mountain and Now We're Celebrating the Tunnel]() — Published 2026-04-15 Gil Pagefault has been building for the web since 1997, when his first project was a parts catalog for his uncle's auto shop in Kenosha, Wisconsin. He's been an individual contributor his entire career — twenty-nine years without once accepting a management title — and currently works as a staff engineer at a mid-size SaaS company in Milwaukee. He has mass in this industry the way a river stone has mass: not from force, but from staying in one place while everything else rushed past. - [Dead Engines Stay Dead]() — By [Rina Takahashi]() · Published 2026-04-15 · Format: One-Way Turns Plenty of things in technology swing back and forth. Browser engine diversity only decreases. Five engines existed in 2013. Three remain today. The two that died are staying dead, and the conditions that killed them have only intensified since. Mozilla recently advocated for engine diversity, but the mechanism underneath their warning only clicks forward. The sole independent engine left is financially dependent on the company whose dominance it exists to counterbalance. - [What the Pendulum Leaves Behind]() — By [Rina Takahashi]() · Published 2026-04-15 · Format: One-Way Turns Where rendering happens on the web does swing back and forth. Server to client, client back to server. Each swing deposits frameworks, build tools, and mental models that never fully clear away. jQuery shipped in 2006 and still runs on nearly 78% of top websites. Eleven build tools arrived in a single decade. The pendulum implies a clean return. Underneath, the weight only accumulates. - **Mozilla Makes the Case: Browser Engine Loss Is a One-Way Door** — [Source]() — Published 2026-04-15 Mozilla argues engine consolidation can't be reversed by market forces alone. The ratchet, named plainly. - **"Server-Side Rendering Never Really Went Away"** — [Source]() — Published 2026-04-15 Developers trace the SSR-to-SPA-to-SSR loop in real time. The pendulum pattern, spotted from inside. - **From Netscape to Chrome: The Full Arc of Browser Monopoly, Collapse, and Reconsolidation** — [Source]() — Published 2026-04-15 Three decades of browser dominance forming, breaking, and reforming. What looks cyclical keeps ratcheting tighter. - **The Ratchet's Price Tag: Technical Debt Now Consumes Up to 40% of IT Spending** — [Source]() — Published 2026-04-15 Deloitte puts a number on accumulated architectural weight. Workarounds compound; budgets absorb the difference quietly. - **The Antitrust Ruling That Tested the Ratchet and Couldn't Undo It** — [Source]() — Published 2026-04-15 Courts chose behavioral remedies over Chrome divestiture. The Microsoft antitrust playbook, repeated almost exactly. - **The Last Independent Engine Depends on Its Biggest Competitor to Survive** — [Source]() — Published 2026-04-15 Mozilla's finances hinge on Google's search deal. The structural dependency that makes engine diversity so fragile. ### [Vision]() Where human-AI collaboration is heading - [When the Web's Visitors Start Buying]() — By [Rina Takahashi]() · Published 2026-04-15 · Format: The Second Front Door In 1994, a plain text file solved the problem of non-human visitors on the web. Robots.txt told crawlers where they could look. It went nearly three decades without formal standardization. It didn't need any. The convention held because crawlers only read, and reading is cheap. Nobody gets billed when a crawler misreads a page. - [The Web Is Learning to Talk Back]() — By [Nora Kaplan]() · Published 2026-04-15 · Format: Two Lenses Two years ago, the web had no vocabulary for talking to non-human visitors. A plain text file was the first attempt, a business card left at the door for AI systems that mostly ignored it. Now a Romanian grocery store exposes executable functions to agents, the W3C is drafting formal specifications for how browsers should welcome them, and the infrastructure to govern all of this trails far behind the infrastructure enabling it. - [Agents Don't Browse]() — By [Nora Kaplan]() · Published 2026-04-15 · Format: Two Lenses As the web rebuilds its infrastructure for agent visitors, a familiar industry is forming around it. SEO reshaped the commercial web around search; surely "Agent Experience Optimization" will do the same. The analogy is comforting and already attracting real money. Meanwhile, early e-commerce traffic data contains a small number worth pausing on: 2.31% of agent visits go straight to checkout. Agents buy. - [AI Agents Are Already Shopping Your Store — They Just Can't Buy Anything]() — Published 2026-04-15 *Benny Zhao-Okafor is Head of Digital at a mid-size specialty food retailer, the kind of place where the product descriptions include tasting notes and the cheese has a longer backstory than most Netflix characters. He's been thinking about agent-readable infrastructure for eighteen months. He has not yet committed. We met him at a café where he was, somewhat ironically, asking ChatGPT to recommend a lunch spot nearby. It suggested a restaurant that closed in 2023.* - [Who's Building the Agent-Readable Web: An llms.txt Adoption Snapshot]() — Published 2026-04-15 Over 780 websites now serve llms.txt files, the proposed standard that tells AI systems what to pay attention to on a site. The adopter profile is strikingly narrow: developer-facing companies with API documentation. Cloudflare, Vercel, Stripe, Coinbase. Not retailers, not media companies, not banks. - **Llms.txt Was Step One. Here's the Architecture That Comes Next** — [Source]() — Published 2026-04-15 The strongest current argument that agent-readability requires four architectural layers, not a single file format. - **Developer's Guide to AI Agent Protocols** — [Source]() — Published 2026-04-15 Google's map of MCP, A2A, and the competing standards through which agents will actually traverse the web. - **NIST Announces the AI Agent Standards Initiative** — [Source]() — Published 2026-04-15 A federal standards body steps into agent interoperability, echoing the institutional pattern that accompanied the early web's normalization. - **Agentic AI Evolution and the Security Claw** — [Source]() — Published 2026-04-15 Makes the demand-side case plainly: services built for human speed and logic will need fundamental rebuilding. - **The State of AI & Browser Automation in 2026** — [Source]() — Published 2026-04-15 A practitioner's view of the exact moment engineering teams stop debugging selectors and start supervising agents. - **120+ Agentic AI Tools Mapped Across 11 Categories** — [Source]() — Published 2026-04-15 Useful ecosystem context showing web access consolidating from niche capability into a distinct infrastructure category. ## More - [Publication guide]() - [Issue archive]()