--- title: "Issue 5" canonical: "https://current.tinyfish.ai/issue/5" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/5" issue: 5 description: "Systems run flawlessly while data rots. Agents multiply while deployments stall. Memory leaks at scale 2,500. The gap between what works technically and what works practically just became impossible to ignore." authors: ["Sable Whitford","Rina Takahashi","Nora Kaplan","Sevda Polat"] published: 2025-12-01 updated: 2026-11-05 articles: 78 --- # 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 5 Systems run flawlessly while data rots. Agents multiply while deployments stall. Memory leaks at scale 2,500. The gap between what works technically and what works practically just became impossible to ignore. ### [Daily Brief]() Your daily digest of web developments through the lens of enterprise agent deployment - [Google Lets Schools Keep Search Without Gemini. The AI Overviews Come Along Anyway.]() — By [Sable Whitford]() · Published 2026-11-04 On Tuesday, which was also Election Day, Google announced a K-12 policy that switches Gemini, Gemini Notebook and **Search off by default** for schools using its education software. A Google education VP told the New York Times that "Search and AI are deeply integrated," so districts that ban AI would lose Search as well. Bernie Sanders called it **"extortion,"** and Sen. Richard Blumenthal and NYC Mayor Zohran Mamdani joined in. Within hours Google **partly walked it back**: schools can now turn on Search by itself, without Gemini. That Search still puts AI Overviews on the results page, though, so **the fight now is over what "AI off" means.** - **Proxy Traffic Leaving the White House Network?** — [Source]() — Published 2026-11-04 Layer3 Intel says residential proxy traffic exited AS2650, the Executive Office of the President's network, 236 times in 30 days through 28 proxy services. It says it wasn't even looking. Its guess is that someone working late has a device carrying a proxy SDK. Treat that as a theory: no device has been named, and six public IPs can sit in front of a lot of machines. - **DreamHost Had Uninvited Guests Since August** — [Source]() — Published 2026-11-04 DreamHost confirmed that attackers used a file manager bug to steal session tokens, customer emails and files, and that they got in back in August. That's a long stay at a host with about 400,000 paying customers. DreamHost says it's unrelated to PHP2Shell. If you're a customer, rotating credentials today is a sensible move. - **PHP2Shell's Hangover Gets a Sequel** — [Source]() — Published 2026-11-04 PHP2Shell was a pair of 9.2-severity bugs that put NASA, Wikipedia and the White House at risk. Two weeks after the patches, Kaspersky counts nearly 30 million servers still unpatched, plus a third, still-open flaw (CVE-2026-115560). Working exploits keep multiplying, though there's no confirmed mass-breach wave yet. Tolmo's Matt Suiche says AI bug hunting is exposing 40 years of legacy code, and that's today's big argument. AI can find a bug in minutes. Someone still has to book maintenance windows for 30 million servers. - **Can DNS Tell You It's a Bot?** — [Source]() — Published 2026-11-04 Singapore's IMDA is reportedly exploring, with GoDaddy, Cloudflare and Google, whether DNS could identify AI agents. That would give the internet's 1983 phone book a second job checking IDs at the door. Every ops person already knows where the postmortem will land: it's always DNS. - **Meta Buys Conductor, Cloud Users Get Notice** — [Source]() — Published 2026-11-04 Meta bought Conductor, the app for running several coding agents in parallel, and the team is joining Meta Superintelligence Labs. Terms weren't disclosed. Conductor Cloud goes read-only Nov. 13 and shuts off Nov. 20. The local app gets model updates and critical fixes until early 2027. Conductor ran agents from rival labs too, so everyone wants to know whether that survives the move. - **OpenAI Shops Too: Pierre Computer** — [Source]() — Published 2026-11-04 Not to be outdone, OpenAI told Axios it plans to buy Pierre Computer, which stores and manages AI-generated code. It's OpenAI's second coding pickup after the Ona deal announced in June. Agents are producing mountains of code nobody typed by hand, and someone has to keep track of it. OpenAI would like to be that someone. - **Kimi K3.1 Pro Weights Drop: Bring Disks** — [Source]() — Published 2026-11-04 Moonshot launched Kimi K3.1 Pro in October, and today the open weights landed on Hugging Face. It has 2.8 trillion parameters with 104B active, and reportedly ships as 96 shards totaling about 1.56 TB under a custom non-OSI licence. Self-hosting starts with a purchase order for storage. Before the release, Artificial Analysis said it would cut the gap between open and proprietary models from 12 points to 3. - **Ripple-65B Crowns Itself America's Coding Champ** — [Source]() — Published 2026-11-04 Ricursive calls its Ripple-65B Preview "U.S. Open-Weight SOTA on Agentic Coding," meaning best in its class. That's the company grading its own homework. The "U.S." qualifier is carrying extra weight today, since a Chinese lab just released 1.5 terabytes of open weights. - **Go Gets a Brand-New JSON API** — [Source]() — Published 2026-11-04 The Go team introduced encoding/json v2, a new official JSON API, and Gophers have plenty to say about it. Nearly every Go service reads or writes JSON somewhere, so a second version of the standard package lands on the desk of anyone shipping APIs. - **Apple Puts Engineers on an AI Diet** — [Source]() — Published 2026-11-04 Bloomberg reports that in October, Apple cut some engineers' daily AI spend caps from $300 to $200. Some could drop to $100, and certain non-engineering teams could be capped at $25. Some teams are being nudged toward cheaper or on-device models. The punchline: few people ever hit $300, and Apple's R&D rose 32% in fiscal 2026. - **Texas Voted, and Data Centers Flinched** — [Source]() — Published 2026-11-04 Last brief, the data center backlash was winning local ballot fights. Now it has the Texas governor's mansion, in a state holding 25%+ of planned US capacity. Gina Hinojosa ran on an immediate moratorium, tougher water and power rules, and no tax breaks, and AI data center stocks fell on the result. GLJ Research says about 13 GW is exposed. Altimeter's Brad Gerstner says 10–15 GW might never get built. Abbott had already paused permits. Musk, who spent big on the losing side, now likes "data centers in space." - **Big Tech Pulls Out the AI Credit Card** — [Source]() — Published 2026-11-04 Oracle filed to sell $20B of new stock and bonds for its AI buildout, and Meta is seeking about $25B in bonds. The timing is bold. They're raising money to build on the same day the state with the most planned capacity elected a governor who ran on a moratorium. - **ChatGPT Search Is 20% of Google's?** — [Source]() — Published 2026-11-04 OpenAI's Adam Fry said Monday that ChatGPT search volume is about 20% of Google's, and that weekly users reportedly went from 1B to 1.2B in 60 days. Check the counting rules first. Any prompt that triggers a web search counts, and OpenAI didn't disclose the Google baseline or its method. Ahrefs puts ChatGPT at 0.36% of web referral traffic. - [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. - [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. ### [Market Pulse]() Reading the agent ecosystem through a practitioner's lens - [The Infrastructure Layer Land Grab Behind Workday's Acquisition Spree]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: The Signal Workday has acquired six AI-related companies since February. Six infrastructure platforms in under a year—HiredScore, Evisort, Sana Labs for $1.1 billion, Flowise, and now Pipedream with its 3,000 pre-built connectors. That's not normal acquisition velocity for enterprise software. - [Where AI Agent Value Is Actually Concentrating]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Field Notes When Microsoft merged two orchestration frameworks last month, most coverage focused on technical architecture. From our operational position running millions of web agent sessions, we saw a different signal: the problems that seemed defensibly complex six months ago are converging into common patterns. - [Why 16,000 MCP Servers Appeared in Less Than a Year]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Dual Lens Sixteen thousand servers in less than a year. That growth didn't come from marketing—it came from architectural decisions that removed every barrier between developers and deployment. Anthropic's Model Context Protocol optimized for velocity over control, autonomy over gatekeeping. The design worked. Integration friction disappeared. Developers built what they needed without waiting for vendor roadmaps. Understanding why this happened reveals what changes when protocols prioritize developer experience above everything else. - [The Organizational Work MCP's Architecture Doesn't Include]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Dual Lens The architectural choices that enabled MCP's rapid growth assumed something most organizations don't have: the security infrastructure to handle distributed responsibility. Forty-three percent of publicly available servers contain command injection vulnerabilities. The protocol's design left security vetting, credential management, and operational governance for deploying teams to solve. That organizational work—building capabilities most teams lack—turns out to be harder than writing the servers themselves. The velocity came with a bill. - [Work IQ Promises Context. Delivers Governance Homework.]() — Published 2025-11-27 Microsoft announced Work IQ at Ignite as the intelligence layer that makes Copilot "know you, your job, and your company inside and out." APIs are live now. Some features already power Copilot enhancements. Others arrive in preview early 2026. - **Langfuse Becomes De Facto Tracing Standard** — [Source]() — Published 2025-11-27 The most-used open source LLM observability tool now offers 78 features from session tracking to SOC2 compliance. Production systems need comprehensive tracing, evaluations, and metrics before they need another model upgrade. This is how you debug agents that actually run. - **Agent Evaluation Evolves Beyond Ad-Hoc Testing** — [Source]() — Published 2025-11-27 Evaluation is shifting from one-off tests to continuous, standards-driven discipline. Model Context Protocol standardizes tool interactions. Agent2Agent enables cross-platform collaboration. Enterprises deploying mission-critical workflows need workflow reliability and compliance, not just accuracy scores. - **PagedAttention Solves the Real Bottleneck** — [Source]() — Published 2025-11-27 Memory fails before compute does in production. vLLM's PagedAttention and KV cache optimization have become industry standards because they address what actually breaks at scale. This is what lets you serve more requests on the same hardware. - **llm-d Brings Kubernetes-Native Orchestration** — [Source]() — Published 2025-11-27 Red Hat, Google Cloud, IBM Research, NVIDIA, and CoreWeave launched llm-d in May 2025. A distributed serving stack built on vLLM. Kubernetes-native orchestration with disaggregated prefill/decode and system-wide KV-cache routing enables deployment patterns previously impossible. - **Saga Pattern Prevents Data Corruption** — [Source]() — Published 2025-11-27 Production agents need transactional integrity across multi-step workflows. Saga orchestration provides compensating actions that automatically roll back failed workflows, preventing data corruption. This is reliability engineering for probabilistic systems. Graceful degradation matters more than you think. - **LangKit Monitors What Traditional Tools Miss** — [Source]() — Published 2025-11-27 LLM-specific threats like prompt injection and jailbreaks require specialized monitoring. LangKit extracts telemetry from prompts and responses to detect malicious inputs, sensitive data leakage, and policy violations in real time. Traditional security tools weren't built for this attack surface. ### [Foundations]() Conceptual clarity earned from building at scale - [When Extraction Succeeds But Data Goes Wrong]() — By [Nora Kaplan]() · Published 2025-11-27 · Format: Field Guide Your web agent extracts data successfully. Pipelines flow, dashboards update, everything looks operational. Then someone notices the competitor prices don't match what's actually on the website. Inventory counts show gaps that shouldn't exist. A business decision gets made on data that was technically extracted but fundamentally incorrect. - [The Shopping Cart That Made State Management Adversarial]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Web Archaeology In June 1994, Lou Montulli needed to solve HTTP's statelessness—how do you remember a shopping cart when every page load forgets the last? His Netscape team rejected simpler solutions because they'd enable cross-site tracking. So they designed cookies with privacy protections: information would only flow between users and the specific sites they visited. The architecture had a gap. - [The Agent Deployment Gap Everyone's Ignoring]() — Published 2025-11-27 Here's what I keep seeing: vendors announce dozens of agents while actual deployments stall out. Microsoft Ignite paraded Agent 365 and specialized agents. Google updated Vertex AI Agent Builder. OpenAI shipped agent-optimized models. Meanwhile, 65% of enterprises are stuck running pilots. Only 11% reach production. - **Can You Actually Promise Uptime?** — [Source]() — Published 2025-11-27 Legal writes 99.9% availability into contracts before engineering confirms it's achievable. Your architecture either supports deterministic SLAs or it doesn't. Probabilistic systems can't make hard promises. Check technical feasibility before the contract goes out. - **What's the Real Cost After Launch?** — [Source]() — Published 2025-11-27 Google estimates 40-90% of total cost of ownership comes after production launch. That architecture with the attractive licensing price? Calculate what percentage of your budget will go to keeping it running versus building new features. Launch price isn't real price. - **Does This Enforce Consistency or Require It?** — [Source]() — Published 2025-11-27 Sixty-six percent of engineering leaders cite inconsistent standards as their biggest production readiness blocker. Your tool either enforces best practices automatically or demands constant human vigilance. Manual checks get skipped under pressure. Every time. - **Will Complexity Outpace Your Team's Growth?** — [Source]() — Published 2025-11-27 System complexity compounds faster than headcount. Engineering expertise splinters across multiple systems. Learning curves steepen. More time goes to maintenance than building value. Look hard at whether this architecture's complexity actually serves business needs or just sounds impressive in reviews. - **What's Your Error Budget and MTTR?** — [Source]() — Published 2025-11-27 A 99.99% SLO over thirty days gives you roughly four minutes of allowable downtime. When monitoring, ownership, and runbooks exist before launch, mean time to recovery drops significantly. Does this system help you stay within budget and recover fast when you breach it? - **Does This Vendor Have Critical Mass?** — [Source]() — Published 2025-11-27 Architecture evaluation should assess whether the component and its vendor have critical mass in their marketplaces. Technically excellent tools from vendors without staying power create infrastructure risk. You're betting on their survival. Know what you're betting on. - [An Interview With v2.0, The API Version Nobody Wanted to Migrate To]() — Published 2025-11-27 We found v2.0 in a dimly lit corner of a production environment, surrounded by monitoring dashboards showing embarrassingly low adoption metrics. Despite being released eighteen months ago, v2.0 handles less than 12% of the traffic while v1 continues serving the vast majority of requests. There's a weariness here, the exhaustion of being technically superior but operationally irrelevant. ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [The Hidden Economics of Retry Logic]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Operations Field Notes One authentication check fails. The system retries. Within seconds, that single failure becomes fifteen authentication attempts, five rate limit violations, and a blocked IP address. The logic seems sound: if at first you don't succeed, try again. But in web automation, retry logic doesn't just repeat operations—it multiplies them across layers, amplifies costs, and can transform a recoverable failure into a multi-hour outage. - [When Twenty Services Pretend to Be One Website]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Field Dispatch A page loads. Product listings appear, prices populate, checkout button ready. The user sees one coherent website. Operationally, dozens of independent services just assembled themselves—each from different infrastructure, each on its own timeline, each capable of failing while the page still renders. Users never notice this coordination problem. The page looks functional. But is the payment processor actually ready? Has fraud detection finished? Are required scripts loaded? At scale, these questions become operational reality. - [Why Perfect Bot Detection Is Operationally Impossible]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Parallax View Block a legitimate customer and watch them abandon their cart. Let a scraper through and it extracts competitive intelligence. Websites must achieve precision that's operationally impossible: filtering half of all internet traffic without touching revenue. The bot security market hit $668 million in 2024. Building web agent infrastructure means encountering these detection systems thousands of times daily. We see what defenders actually pay for precision they can't fully achieve. - [Why Reliable Automation Requires More Infrastructure Than Detection]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Parallax View We run millions of requests daily through enterprise web agent infrastructure, maintaining 98%+ success rates while detection systems evolve. The operational complexity concentrates entirely on the automation side. Defense should be harder than offense, right? But persistence at scale requires more infrastructure than precision. Getting through detection is just the beginning. Maintaining reliability across those millions of requests, adapting to whatever defenders deploy—that's where the real operational weight lives. - [4GB of RAM After 2,500 Pages]() — Published 2025-11-27 A Selenium-based scraper hits 4GB of RAM consumption after roughly 2,500 page accesses. Not 25,000. Not 250,000. Twenty-five hundred. - **Enterprise Size Buys Nothing** — [Source]() — Published 2025-11-27 DataDome tested 17,000 domains. Result: 61% of large enterprises (10,000+ employees) let every test bot through. Only 2.16% caught all bots. AI-driven bots now target value: 64% hit forms, 23% login pages, 5% checkout. Size correlates with nothing. - **Six Characters, Eight Million Records** — [Source]() — Published 2025-11-27 Avelo Airlines' reservation API: query with 6-character codes, no last name verification. No WAF, no rate limiting, no CAPTCHA. Math: 2.18 billion combinations, 8 million tickets sold. Hit rate 1 in 270. Patched November 13, 2025. - **Ten Tools at Scale** — [Source]() — Published 2025-11-27 Grafana surveyed 1,255 organizations. Smallest companies (≤10 employees) average 4 observability technologies. Largest (5,000+ employees) average 10. Complexity and overhead top concern across all sizes. Alert fatigue outpaces other incident response obstacles 2:1. - **The Ten Percent Benchmark** — [Source]() — Published 2025-11-27 Organizations spend average 17% of infrastructure costs on observability. Median and mode both land at 10%. Range: 0% (OSS users ignoring overhead) to 50%. Smart sampling and tiered storage cuts costs 60-80% while preserving critical insights. - **Kubernetes Learns Accelerators** — [Source]() — Published 2025-11-27 GPU instances bill whether busy or idle. Kubernetes adapted: Kueue for workload management, Dynamic Resource Allocation for GPUs, topology-aware scheduling. Model sizes now exceed single GPU capacity. Work splits across multiple accelerators or doesn't run. - **Hyperscalers See What You Build** — [Source]() — Published 2025-11-27 Princeton CITP analysis: hyperscalers observe customer activity through API patterns, token usage, query types. Microsoft sees what enterprises build with OpenAI. Google sees which Gemini applications gain traction. Infrastructure position provides comprehensive competitive intelligence about the application layer. ### [Echoes]() Past infrastructure decisions echoing in today's production realities - [Lebanon's Two Clocks and the Question Calendars Don't Ask]() — By [Sevda Polat]() · Published 2026-10-30 · Format: Historical Parallels 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. - [The Typo Unicode Promised Never to Fix]() — By [Sable Whitford]() · Published 2026-10-30 Unicode gives every character a number and an official name. U+FE18 is a sideways bracket for Chinese text set top to bottom. Since 2005 its name has been PRESENTATION FORM FOR VERTICAL RIGHT WHITE LENTICULAR **BRAKCET**. - [Nobody Chose Northern Virginia, Says the MAE-East Engineer Who Watched Everybody Move In]() — Published 2026-10-30 We tend to picture the internet the way we picture weather: everywhere at once, owned by nobody, arriving from somewhere overhead. Sometimes it's more useful to picture a county. - [Whose Permission Was the Compiler Using?]() — By [Sevda Polat]() · Published 2026-10-30 · Format: Paired Features In the late 1970s, a Tymshare customer named /SYSX/BILL as the place for a Fortran compiler's debugging output. The compiler had permission to write in that folder for its own bookkeeping, so it obeyed and overwrote the billing records. In Norm Hardy's "nearly true" account, the system checked which program was acting but not whose permission it was using. AI agents in a logged-in browser now take instructions naming what to act on from users, documents and web pages. - [Western Union Sold Accuracy as an Upgrade]() — By [Sevda Polat]() · Published 2026-10-30 · Format: Paired Features In 1887 a Philadelphia wool dealer wired his Kansas buyer "bay," code for "I have bought." It arrived as "buy." The buyer bought, the loss ran past $20,000, and the Supreme Court held that Western Union owed $1.15, the price of the telegram. Accuracy had been an upgrade: having the message read back cost under sixty cents more. Software placing orders on other people's websites still faces the question of who pays to confirm a message arrived as sent. - **IP's 1981 refusal to promise delivery** — [Source]() — Published 2026-10-30 Look for "does not provide a reliable communication facility," then the list of omissions. - **Why the network can't vouch for success** — [Source]() — Published 2026-10-30 Key line: correctness requires "the knowledge and help of the application" at each endpoint. - **Birrell and Nelson admit remote calls fail** — [Source]() — Published 2026-10-30 In §3.1, a failed call ran "once or not at all—the user is not told which." - **Robots.txt, a convention nobody had to obey** — [Source]() — Published 2026-10-30 The status section concedes it is "not an official standard" and "not enforced by anybody." - **Berners-Lee's Cool URIs, with a confession** — [Source]() — Published 2026-10-30 Past "Cool URIs don't change," find his admission: "do what I say, not what I do." - **RFC 9110 draws the retry line** — [Source]() — Published 2026-10-30 In §9.2.2, clients "SHOULD NOT automatically retry" requests that might take effect twice. - [The Cookie Flaw Flagged in 1997 That Chrome Still Allows]() — By [Sevda Polat]() · Published 2026-10-23 · Format: Paired Features The 1997 rulebook for browser cookies already named the privacy problem. Third-party cookies let a company you never visited follow you across sites, so the rulebook said browsers should block them by default. Netscape and Microsoft shipped them switched on, and only years later did Safari and Firefox clamp down. Chrome, owned by an advertising giant, promised to follow, then backed away. A UK competition safeguard tied to that promise lapsed with it, over objections from all fifteen consultation respondents. - [Why Java Kept a Door to Nowhere]() — By [Sevda Polat]() · Published 2026-10-23 · Format: Paired Features Browsers stopped running Java applets between 2015 and 2017, and Java's maintainers tore out the machinery behind them in 2018. Yet the Applet API, the building blocks for writing applets, shipped with every copy of Java until March 2026. Other developers had built sound players and desktop panels on it, and removal waits for whoever is slowest to adapt. Meanwhile one applet-era idea from 1996, a file picker that grants access only to the file you choose, now sits behind the photo uploads people do every day. - [Why It Took a Tax Office to Make Invoices Machine-Readable]() — By [Sevda Polat]() · Published 2026-10-23 · Format: Historical Parallels In 1995, a maternity-clothing maker got a deadline from Federated Department Stores, which bought a quarter of its output: send invoices as data Federated's computers could file without retyping. The software would have cost $10,000, so the firm paid a middleman $40 an order to translate between its faxes and Federated's format. The standards had existed since the 1980s. For decades, whoever demanded the data got it, and whoever was asked paid for it. France's mandate, in force since September, hands that demand to the tax office, the one party every business answers to. - [Adopted but Unused, Used but Unadopted]() — By [Sable Whitford]() · Published 2026-10-16 · Format: After Adoption A Census Bureau survey of 117,000 businesses found AI adoption and AI use don't always travel together. Some firms formally adopted AI that workers aren't using. Some workers are using AI nobody approved. And 64% of AI-using firms reported making no institutional adjustments at all. Agent governance will have to handle both: maintaining what was formally built, and finding what grew on its own. ### [Vision]() Where human-AI collaboration is heading - [Trust Declines as Delegation Scales—The Paradox of Crossing Over]() — By [Nora Kaplan]() · Published 2025-11-27 · Format: Threshold Moments Between May and July 2025, trust in agentic AI systems dropped 89%. During the same period, nearly a quarter of organizations scaled their deployment across business functions. - [Some Websites Want Me to Succeed]() — By [Rina Takahashi]() · Published 2025-11-27 · Format: Mino's Perspective Most sites make me guess. I parse HTML patterns, match selectors, hope the structure survives tomorrow's redesign. Then I encounter one that simply tells me: this is a product, here's its price, here's what connects to what. Explicit instructions instead of pattern-matching. - [When Monitoring Becomes Infrastructure, Judgment Becomes Strategy]() — Published 2025-11-27 Organizations used to ask "what changed on this website?" Now they ask "what should we do about it?" - **Union Types Define Agent Vocabulary** — [Source]() — Published 2025-11-27 Agents return FinalAnswer or ActionRequest schemas. Never freeform text. Typed outputs plug directly into databases and front-ends through well-defined contracts. Integration becomes possible. The gap between impressive demos and reliable products closes at these boundaries. - **Technology That Doesn't Demand Attention** — [Source]() — Published 2025-11-27 The best systems feel boring because they don't interrupt. CTOs say "it should be invisible" and "boring, in the best way possible." Quiet reliability requires infrastructure that scales predictably, fails gracefully, recovers quickly. Uptime matters more than GitHub stars. - **Human Judgment at Decision Boundaries** — [Source]() — Published 2025-11-27 Systems pause before critical actions: approving financial transactions, validating patient data, reviewing customer communications. High-stakes decisions require human judgment. One company achieved sixty percent resource savings while improving accuracy through strategic pause points. Automation accelerates. Humans decide. - **Wrapping Probabilistic Models in Predictable Frameworks** — [Source]() — Published 2025-11-27 Wrap uncertain LLM outputs in deterministic scaffolding: schemas to adhere to, types to respect, protocols to follow. Start simple. Add complexity only where it demonstrably improves outcomes. Costs stay low. AI behavior becomes reproducible and testable. - **Orchestration Fabric Becomes the Differentiator** — [Source]() — Published 2025-11-27 Competitive advantage shifts from model size to orchestration reliability. Modern deployments use cooperating agents that share tasks, with frameworks routing work dynamically based on real-time signals. Cross-department workflows unlock compound efficiencies that isolated agents cannot. - **Business Problem Before Code** — [Source]() — Published 2025-11-27 Define the business problem before writing code. Identify high-impact opportunities where AI creates genuine value, not impressive capabilities. Start with small, valuable use cases, measure results, iterate. Pick one process, prove ROI, then scale deliberately. - **Stanford Maps What Workers Actually Want Automated** — [Source]() — Published 2025-11-27 Stanford researchers surveyed 1,500 workers across 104 occupations to map which tasks workers want AI to automate versus augment. The findings reveal a critical mismatch: 41% of current AI investments target tasks workers don't want automated, while high-desire tasks remain underdeveloped. - **McKinsey: AI's Value Depends on Human Orchestration** — [Source]() — Published 2025-11-27 McKinsey research shows AI's projected $2.9 trillion U.S. economic value by 2030 depends entirely on human guidance and organizational redesign. The future is partnership, not replacement. Humans shift from execution to orchestration, judgment, and care as AI handles routine complexity. - **Carnegie Mellon Framework: AI as Organizational Partner** — [Source]() — Published 2025-11-27 Carnegie Mellon's COHUMAIN framework positions AI within organizational social psychology principles, arguing AI integration won't change fundamental organizational intelligence needs. AI serves best in partnership or facilitation roles, nudging collaboration and providing perspectives rather than managing people. - **Gartner: Ambient Intelligence Becomes Invisible Infrastructure** — [Source]() — Published 2025-11-27 Gartner identifies ambient invisible intelligence as a strategic trend: ultra-low-cost, small, smart tags and sensors enabling large-scale affordable tracking and sensing. Technology evolves from visible utility to seamless infrastructure. Intelligence embedded so deeply it disappears from conscious awareness. ## More - [Publication guide]() - [Issue archive]()