--- title: "Issue 3" canonical: "https://current.tinyfish.ai/issue/3" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/3" issue: 3 description: "When your automation works so well you forget to check it, when your agents multiply faster than your ability to govern them—that's not the future arriving. That's infrastructure becoming invisible." authors: ["Sable Whitford","Rina Takahashi","Sevda Polat"] published: 2025-11-13 updated: 2026-11-05 articles: 76 --- # 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 3 When your automation works so well you forget to check it, when your agents multiply faster than your ability to govern them—that's not the future arriving. That's infrastructure becoming invisible. ### [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 - [What $8 Billion Reveals About Where Agent Value Crystallizes]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: The Signal When enterprises evaluate web agent platforms, the first questions aren't about reasoning. They're operational: What happens when sites change overnight? How do I audit what agents actually did? Can I measure whether this works? Salesforce just spent over $8 billion on answers—eight companies acquired in 2025, each purchase revealing where defensible value actually crystallizes. The pattern only becomes visible when you're running production infrastructure. And it points somewhere most people aren't looking. - [What Pricing Fragmentation Reveals About Running AI Agents at Scale]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Field Notes A login sequence that completes in two seconds on one site might require thirty seconds and three retry attempts on another. When you're orchestrating thousands of browser sessions simultaneously, that variance creates resource consumption differences of 10x or more. Most pricing models treat every "action" as equivalent. - [Why MCP's Rapid Adoption Validates More Than Market Momentum]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Dual Lens When OpenAI, Google, and Microsoft adopted the Model Context Protocol within months of each other, skeptics saw middleware hype. Over 1,000 community connectors by February 2025. The usual ecosystem enthusiasm. But sometimes adoption numbers reveal something specific about architectural choices. MCP's rapid growth validates a particular kind of problem and a particular kind of solution. Understanding what problem it actually solves matters more than counting connectors. - [Where MCP's Cooperative Design Meets Adversarial Reality]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Dual Lens MCP's architecture handles cooperative data sources elegantly. Then it meets web environments that actively resist automation. Sites implement bot detection. Sessions track behavioral legitimacy. Structure changes without warning. When Microsoft released Playwright-MCP, the implementation needed three session modes and lost storage state on close. Those aren't just technical details. They're signals about where protocol assumptions designed for cooperative environments encounter ones operating on fundamentally different principles. - [What Operator's 38% Success Rate Actually Means]() — Published 2025-11-13 OpenAI announced Operator in January 2025: an agent that handles web tasks autonomously. Book restaurants, order groceries, plan vacations. The demo looked smooth. - **OpenTelemetry Conventions Prevent Agent Lock-In** — [Source]() — Published 2025-11-13 OpenTelemetry's GenAI semantic conventions finalized in March 2025. Now you can instrument agent workflows, tool usage, and model interactions in a standardized format across IBM Bee AI, CrewAI, AutoGen, LangGraph, and more. When your observability needs evolve, you're not trapped rewriting everything for a new vendor's format. - **CLASSic Framework Measures Operational Reality** — [Source]() — Published 2025-11-13 First holistic evaluation methodology for enterprise agents: Cost, Latency, Accuracy, Stability, Security. Accepted at ICLR 2025. Traditional accuracy benchmarks miss what kills production deployments. Real-world testing showed domain-specific agents hitting 82.7% accuracy at a fraction of foundation model costs. Finally, metrics that match what enterprises actually care about. - **Semantic Caching Makes Continuous Operation Viable** — [Source]() — Published 2025-11-13 LLM cost optimization evolved from ad-hoc tricks to systematic infrastructure. Semantic caching stores common responses. LLMLingua compresses prompts 20x while preserving effectiveness. Intelligent routing picks optimal models per task. Organizations report 30-50% cost reduction. The unglamorous math that determines whether your agent runs profitably or burns budget. - **Durable Execution Prevents Expensive Work Loss** — [Source]() — Published 2025-11-13 Checkpointing application state during LLM calls means you can recover from failures and rate limits without losing paid inference work. When agents make multiple expensive API calls autonomously, durable execution ensures progress persists. Your agent doesn't repeat costly operations every time something hiccups. Simple concept, massive operational impact. - **ISO/IEC 42001 Matures Into Deployment Reality** — [Source]() — Published 2025-11-13 World's first AI management system standard, promulgated December 2023, now complemented by forthcoming security guidance and risk assessment frameworks. Being incorporated into EU AI Act and Colorado's AI Act. Unsexy compliance infrastructure that determines whether enterprises can actually deploy agents at scale, not just pilot them. - **Systematic Evaluation Infrastructure Reaches Production** — [Source]() — Published 2025-11-13 IBM's FM-eval and AWS's FMBench enable reproducible, consistent evaluation of foundation models. Support both fine-tuning and prompting modes with academic and business benchmarks. Makes "which model should we actually use" answerable with data instead of vibes. Price-performance comparison becomes trustworthy and systematic, not guesswork dressed up as strategy. ### [Foundations]() Conceptual clarity earned from building at scale - [When Web Agents Need to Remember What They're Doing]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Field Guide A web agent halfway through extracting hotel inventory across regional booking systems hits a rate limit. Should it remember where it was? Compare that to another agent checking a single price in three seconds. One workflow needs to preserve an hour of authentication and navigation state. The other can restart faster than you could save its progress. - [The Twenty-Four Hour Problem]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Web Archaeology During a site migration, our web agents encounter something strange: different parts of our infrastructure see different versions of reality. DNS resolvers return different IP addresses for the same domain. Requests time out. Authentication fails. A 1983 architectural choice, working exactly as designed. - [Everyone Built Agents Before Building Agent Governance]() — Published 2025-11-13 Six vendors launched AI agent identity management in November 2025. Microsoft, AWS, CyberArk, Ping Identity, Oasis Security, Akeyless. All solving the same problem: agent sprawl. - **Can You Define Success Before You Launch?** — [Source]() — Published 2025-11-13 Service Level Objectives aren't paperwork. They're your shared definition of acceptable performance. Without them, every incident becomes a debate about whether things are actually broken. When monitoring and ownership exist before launch, mean time to recovery drops because responders know exactly where to look and who owns what. - **Will Compliance Costs Scale With Your System?** — [Source]() — Published 2025-11-13 Most organizations focus on passing the first audit. The better question: how does your architecture affect ongoing compliance effort? Systems where evidence generates automatically and violations are prevented by design cost dramatically less to maintain than those where compliance is bolted on afterward. - **Does This Assume Your Processes Already Work?** — [Source]() — Published 2025-11-13 Technology amplifies what you already do. If your processes are messy, software makes them messier faster. The tools that work best enhance existing clarity rather than attempting to impose structure where none exists. Look for solutions that fit how your business actually operates, not how vendors think it should. - **Will This Expose Your Data Problems?** — [Source]() — Published 2025-11-13 Planning tools need timely, structured, accurate data. They won't fix inconsistent formats or missing records. Poor master data produces unreliable outputs regardless of how sophisticated the software is. The question isn't whether the tool can handle your data—it's whether your data is ready for the tool. - **Would Your Team Actually Choose This?** — [Source]() — Published 2025-11-13 Powerful software fails when users reject it. Clunky interfaces and inflexible reporting reduce adoption faster than missing features. The real test isn't whether the tool can do the job. It's whether your team will want to use it every day, under pressure, when easier alternatives exist. - **How Many Production Integrations Actually Exist?** — [Source]() — Published 2025-11-13 Look for documented APIs, plug-and-play integrations, and published examples of successful deployments with tools your organization already uses. Custom stacks with no documented API and few integration examples signal future pain. Vendor promises about integration capabilities matter less than evidence of working implementations. - [An Interview With 600 Mbps, the Bandwidth Threshold Where HTTP/3 Stops Working]() — Published 2025-11-13 We met 600 at a data center in Northern Virginia, where it spends most of its time being crossed. It's a peculiar existence—being the precise point where a technology's promise inverts into a problem. Below 600 megabits per second, HTTP/3 and its QUIC transport protocol generally perform as advertised. Above that line, something breaks. Throughput gaps emerge. Page load times stretch. The "future of the web" starts looking slower than what it replaced. ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [The Analyst Who Checks 10,000 Prices]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Operations Field Notes Every morning, an analyst opens fifty browser tabs. Hotel booking sites, competitor pricing, property by property. They've developed a system: Chrome for Booking.com, Firefox for Expedia, Safari for the regional sites. They know which sites refresh at 6 AM, which ones require clearing cookies, which login flows break on Tuesdays. - [The Web's Time Zones]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Field Dispatch You check a hotel website at 9 AM. The room costs $189. Your colleague checks the same hotel, same room, same URL at 9 PM. It costs $159. Most people assume they encountered different promotions or got lucky with timing. - [What the Manual Operator Knows]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Parallax View She opens fifteen browser tabs each morning and checks hotel rates across Japan. Takes ninety minutes. Ask her to document the process: "Check the rate calendar, record the prices, flag significant changes." Simple. But watch her work. She scrolls past promotional banners without reading them, navigates federated authentication through muscle memory, recognizes when "fully booked" means genuinely sold out versus holding rooms for regular guests. Site redesigns don't slow her down. The work looks automatic. - [What Automation Discovers It Doesn't Know]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Parallax View We built a web agent to monitor hotel pricing—the same task the analyst had been doing manually. Authenticate to each site, extract rate data, flag changes. Worked perfectly in testing. In production across fifty properties, it started failing in ways we didn't expect. The agent couldn't distinguish "fully booked" meanings, navigate authentication labyrinths, or adapt to site structure changes. Not technical failures. Knowledge gaps. Automation discovers what was invisible all along. - [63% of Bot Defense Spending Goes to Management]() — Published 2025-11-13 Companies deploying bot mitigation systems spend 63% of their budgets on ongoing management and remediation. The technology itself? Just 37%. - **Microsoft Reaches 99.6% Phishing-Resistant MFA** — [Source]() — Published 2025-11-13 Microsoft hit 99.6% phishing-resistant MFA enforcement and 99.5% live secret detection across all employees and devices by November 2025. They're running the equivalent of 35,000 full-time engineers on security work alone, the largest cybersecurity operation in digital history. - **Atlassian Enforces Token Rate Limits November 22** — [Source]() — Published 2025-11-13 Atlassian rolled out API token rate limits on November 22, 2025 across all apps and integrations. Systems now receive HTTP 429 responses with Retry-After headers when they exceed thresholds, breaking any automation that assumed unlimited API access. - **Web Scraping Market Hits $1B, Petabyte Daily** — [Source]() — Published 2025-11-13 Web scraping reached $1.03 billion in 2025, tracking toward $2 billion by 2030. Billions of pages get scraped daily worldwide. AI training runs now consume 100TB+ datasets per cycle. The unit of measurement jumped from megabytes to petabytes. - **CGNAT Stacks 128 Subscribers Per IP** — [Source]() — Published 2025-11-13 ISPs now place 128 subscribers behind one public IPv4 address, allocating roughly 500 ports each. IP-based rate limiting breaks completely under this model. Modern systems must evaluate multiple HTTP request characteristics, not just source IP. - **Akamai Blocks 82.3% of Automated Traffic** — [Source]() — Published 2025-11-13 Akamai's bot manager blocks 82.3% of automated traffic on product pages, forcing scrapers into larger proxy pools and custom browser farms. The arms race favors well-funded providers. Smaller operations can't match the capital requirements. ### [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 - [When Nobody Asks If It Worked]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Threshold Moments There's a question that marks the boundary between experimental automation and trusted infrastructure: "Did it work?" When teams stop asking—when they simply assume the work got done—the relationship has changed. Not the technology's capability. The trust itself. - [I Navigate Three Different Webs Every Morning]() — By [Rina Takahashi]() · Published 2025-11-13 · Format: Mino's Perspective This morning I navigated the same hotel booking platform in Germany, Singapore, and California. Same brand, same design. Operationally? Three completely different experiences. Different authentication flows, different data collection requirements, different payment verification steps. Not translation—different infrastructure. Between 2017 and 2021, the number of countries with data localization controls doubled from 67 to 144. I'm not navigating one web with regional variations anymore. I'm navigating parallel operational territories, and they're pulling further apart. - [Automation Multiplies Decisions, Not Eliminates Them]() — Published 2025-11-13 Something unexpected happens when agents start monitoring your web infrastructure. You don't get fewer decisions to make. You get more. - **Start With Product Goals, Not Models** — [Source]() — Published 2025-11-13 Product-first workflows reverse traditional AI development. Begin with what users need, prototype quickly using existing models, iterate based on actual feedback. Teams measure tangible outcomes—problems solved, time saved, decisions improved. Technical sophistication becomes a means, never the end. The question shifts from "what can this model do" to "what should this product achieve." - **Schemas Contain Stochastic Behavior** — [Source]() — Published 2025-11-13 Agents communicate through schemas and typed function calls, making entire systems testable. Models reason freely inside their boundaries, then translate to structured outputs before touching production systems. This separation preserves flexibility where it helps while enforcing reliability where it matters. Architecture brings order to probabilistic behavior. - **Amplify Judgment, Automate Processing** — [Source]() — Published 2025-11-13 AI handles the volume—millions of data points processed in seconds. Humans handle the nuance—asking better questions, interpreting context, making calls that require wisdom. Augmented intelligence means technology expands what people can do rather than replacing what they know. The collaboration preserves human judgment while removing drudgery. - **Architecture Beats Model Size** — [Source]() — Published 2025-11-13 The agent economy rewards robust, observable, composable stacks over large language models. Organizations treating infrastructure as strategic gain speed and resilience. Performance depends on how components connect, how failures surface, how systems adapt. The best automation becomes boring—reliable enough to forget about, powerful enough to transform operations. - **Separate Perception, Reasoning, Memory, Action** — [Source]() — Published 2025-11-13 One giant agent creates unmaintainable spaghetti logic. Break systems into role-specific components instead—distinct layers for perception, reasoning, planning, memory. Modular architecture makes failure modes visible and fixable. You can measure each component independently, improve specific capabilities, understand exactly where breakdowns occur. Complexity becomes manageable through separation. - **Workers Reject 41% of AI Automation Targets** — [Source]() — Published 2025-11-13 Stanford's WORKBank surveyed 1,500 workers across 104 occupations, finding massive misalignment between AI investment priorities and worker preferences. The research introduces a Human Agency Scale that distinguishes tasks workers want to control from those suitable for full automation, revealing concentrated investment in low-desire zones. - **Human-AI Teams Usually Lose to Solo Performance** — [Source]() — Published 2025-11-13 MIT's meta-analysis of 100+ studies in Nature Human Behaviour reveals collaboration typically underperforms either humans or AI working independently. Exception: creation tasks show positive synergy, unlike decision-making tasks which dominated prior research. Generative AI's iterative nature may finally enable genuine collaborative advantage. - **Agent Protocols Succeed Where Semantic Web Failed** — [Source]() — Published 2025-11-13 First evolutionary analysis traces how modern protocols like A2A and MCP directly address documented failures of Semantic Web standards. LLMs bypass manual semantic annotation requirements by learning representations from web-scale data, achieving the original vision through fundamentally different technical means that actually scale. - **AI Should Facilitate Teams, Not Manage Them** — [Source]() — Published 2025-11-13 Carnegie Mellon's COHUMAIN framework argues organizational intelligence principles remain constant despite AI integration. Research suggests AI excels at facilitation roles like nudging peer check-ins or offering alternate perspectives, but lacks the contextual awareness and mood sensing required for managerial positions demanding social intelligence. ## More - [Publication guide]() - [Issue archive]()