--- title: "Issue 2" canonical: "https://current.tinyfish.ai/issue/2" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/2" issue: 2 description: "Production reality keeps humbling our demos. Thirteen trillion bot requests blocked. Agents failing at enterprise scale. Detection systems seeing through code. This week: the infrastructure gap nobody wants to talk about." authors: ["Sable Whitford","Rina Takahashi","Sevda Polat"] published: 2025-11-05 updated: 2026-11-05 articles: 86 --- # 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 2 Production reality keeps humbling our demos. Thirteen trillion bot requests blocked. Agents failing at enterprise scale. Detection systems seeing through code. This week: the infrastructure gap nobody wants to talk about. ### [Daily Brief]() Your daily digest of web developments through the lens of enterprise agent deployment - [The Accounts You Locked Were the Ones That Opened: Meet \*sshole]() — By [Sable Whitford]() · Published 2026-11-05 OpenSSH and Alpine Linux **shipped fixes today** for `*sshole`, a decade-old bug that let any password between 257 and 1024 bytes log in on Linux boxes running musl libc without PAM. Locked accounts get a `*` in `/etc/shadow`. musl's `crypt()` also returns `*` when it chokes on a password. OpenSSH compared the two, saw a match, and let you in. The bug was reported Wednesday and patched **this Thursday** with no embargo, because maintainer Damien Miller says AI agents now rediscover bugs within hours. Upgrading alone won't protect you. You still have to restart sshd. - **OpenAI buys the Git plumbing under Codex** — [Source]() — Published 2026-11-05 OpenAI is acquiring Pierre Computer Company, maker of Git-based distributed storage "with API ergonomics designed for machines." Pierre's open-source Trees and Diffs libraries already run file browsing and change review in Codex, and they stay open source. Terms undisclosed. It follows Pando, OpenAI's in-house source-control backend, which reportedly cut deploy latency about 80%. Coverage is calling this a GitHub challenger. OpenAI hasn't said so. - **Meta buys Conductor, users get eviction notice** — [Source]() — Published 2026-11-05 Same day, different buyer: Meta picked up Conductor, the app for running several coding agents at once. Sign-ups and downloads are frozen and the cloud side is winding down. Cloud data deletion starts November 20, so export what you care about now. CEO Charlie Holtz promises something new early next year. Meta hasn't said whether existing integrations survive. - **GitHub picks a terrible day to wobble** — [Source]() — Published 2026-11-05 From 09:49 UTC Wednesday, Git operations, Issues and pull requests degraded, then Actions and the API joined in. Mitigated around 11:30, resolved by 11:50, with a few more Actions hiccups later on. All of it landed the same day OpenAI bought a Git storage company built for agents. Rough optics. Rougher pager. - **Revolut swaps Claude Code for self-hosted GLM** — [Source]() — Published 2026-11-05 About 2,000 Revolut engineers now default to a self-hosted Z.ai GLM model: open weights, Revolut's own US servers, roughly 56 Nvidia B200s, about $0.30 per million tokens, around a tenth of the old bill. Claude Code and Codex stay for the hardest work, and Gemini gets images. Revolut reports 70% better lead time and 60% better coding time. AI ops head David Abitbol says the bottleneck "is not necessarily coding." Cheap and good enough is winning fans, and people are loudly arguing over the stigma around Chinese open weights. - **Rent-a-proxy, now exiting from the White House** — [Source]() — Published 2026-11-05 Layer3 Intel saw residential proxy traffic exit from AS2650, the Executive Office of the President's network, 236 times in a 30-day window ending Tuesday. That was 6 addresses, on 26 of the 30 days, across 28 proxy services. Layer3 wasn't hunting for it. The addresses just turned up in rotation. The pattern looks like someone's everyday device rather than guest Wi-Fi. No breach is shown and the device is unidentified. Sentient, via Cybernews, suspects Rave.io, an app bundling 12 proxy SDKs. The White House hasn't commented. - **CISA flags an actively exploited WordPress flaw** — [Source]() — Published 2026-11-05 CISA added a WordPress Core vulnerability to its Known Exploited Vulnerabilities catalog, so attackers are already using it. Details are thin. Do the boring thing: update WordPress, check that auto-updates actually ran, and poke whoever supposedly "owns" the marketing site. - **Bot swarms are knocking university libraries offline** — [Source]() — Published 2026-11-05 Data-hungry bot swarms are hitting university library websites hard enough to crash them. Sites built for students and researchers are soaking up crawler traffic that will never borrow a single book. Toss it on the scraper-fatigue pile, which keeps getting taller. - **Does Gemini train on paid chats?** — [Source]() — Published 2026-11-05 A complaint making the rounds says Gemini trains on chat history even for paying subscribers. It's an unconfirmed claim that people are arguing about loudly. Either way, it's a decent excuse to go poke at your own settings. - **Nvidia's AI laptop: unplug it to go faster** — [Source]() — Published 2026-11-05 On RTX Spark laptops running GPT-OSS-120B, Windows and WSL hit about 65 tokens per second. Native Linux managed 13.9, roughly 79% slower. Pull the power cord and Linux climbs back to about 61–67. MindStudio blames a Windows-only firmware component with an encrypted power table Linux can't decrypt. That's one study, and Nvidia hasn't confirmed it. Battery mode as a performance feature. Wild. - **AMD expects to outsell Intel, a first** — [Source]() — Published 2026-11-05 Per Tom's Hardware, AMD expects to outsell Intel in Q4 for the first time ever. That's a forecast, so wait for the final count. Still, anyone who remembers AMD as the perennial underdog gets to enjoy that sentence. - **Data centers get voted down, town after town** — [Source]() — Published 2026-11-05 Data centers had a bad night in Tuesday's midterms. Bloomberg headlines say only one town out of dozens voted in favor of them, and AI Weekly counts bans or restrictions passing in at least 14 Ohio municipalities. Wall Street read the night its own way: Bloomberg reports Big Tech rallied as the results reshaped the AI spending outlook. - **Baidu ships ERNIE 5.5 Flash Base** — [Source]() — Published 2026-11-05 Baidu released ERNIE 5.5 Flash Base. For now there's a release page and not much else, so no specs to argue about yet. It lands on a day when Chinese models were already the talk of engineering teams. - **OccupySaaS: laid-off engineers clone everything** — [Source]() — Published 2026-11-05 Under the OccupySaaS banner, laid-off engineers are using AI to build open-source alternatives to, well, everything. Auth0 and Salesforce clones are in the works. Free time plus a code generator turns out to be a hell of a combination. - **AI is calling in 40 years of tech debt** — [Source]() — Published 2026-11-05 An IEEE piece argues AI is forcing organizations to finally pay off four decades of technical debt. The headline alone had people glancing nervously at the oldest box in their stack, the one nobody is allowed to reboot. "We'll fix it later" apparently had a due date. - **A detective game where every suspect lies** — [Source]() — Published 2026-11-05 Uncracked City casts LLMs as your suspects, and they lie to you, on purpose this time. You interrogate them, cross-check their stories and catch the contradictions. A model making things up is the whole point here, and it's suspiciously good practice for reading AI output at work. - **Stringman: a room-sized robot gripper, $800** — [Source]() — Published 2026-11-05 Stringman is an open-source robot gripper that works across an entire room, for about $800. That's roughly used-bike money, which puts room-scale robotics within reach of anyone with a free weekend and a garage. Expect weekend projects. - **MCP Explained, for when the acronym blurs** — [Source]() — Published 2026-11-05 MCP Explained is a standalone site that walks through Model Context Protocol, the standard for hooking AI models up to tools and data. Bookmark it for the next meeting where someone says "just wire it up with MCP" and the whole room nods like they know. - **Download your favourite YouTube videos. Seriously.** — [Source]() — Published 2026-11-05 A warm little essay makes the case for keeping local copies of the YouTube videos you love, because channels vanish and links rot. Ops people will agree on sight: if it matters and there's no backup, you're only borrowing it. - [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. - [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. - [Who Pays While the Agent Waits]() — Published 2026-10-30 Google's new service for hosting AI agents starts from a fact anyone who has watched one run already knew: most of the time, the agent is waiting, for the model to answer or for a person to reply. So Google suspends idle agents and promises you won't pay for idle machines. Fair enough, but the wait still gets paid for. It moves into approval queues nobody times, and then onto the plan, which comes back exactly as it left, even if the stock it counted on has since sold. - [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 Infrastructure Gap Between Promising Outcomes and Charging for Them]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: The Signal Been watching the clouds gather around seat-based pricing for months—every customer conversation circles back to "why am I paying for licenses when agents could do this?" Automation Anywhere buying Aisera this week feels like the first real attempt to actually build infrastructure for charging by outcomes, not just talking about it. Their 40% fewer ITSM seats claim isn't what got me—it's that they're betting they can prove what agents accomplished and charge for that at enterprise scale. - [The Work That Never Was]() — By [Rina Takahashi]() · Published 2025-11-07 A company sits on 50,000 customer contracts. Somewhere in those documents: signals about which customers might buy the next product. Valuable intelligence that could drive millions in revenue. But no one ever reads them. Not because it's too expensive to hire people. Because the work simply never happened. Box CEO Aaron Levie keeps hearing this pattern from enterprise customers deploying AI agents. Everyone thinks they know why. They're wrong about the barrier. - [The Work That Never Was]() — By [Rina Takahashi]() · Published 2025-11-07 A company has 50,000 customer contracts. Mining them for upsell signals would be valuable. They've never tried. Not because they didn't want to—because attempting it would bleed money at any price. Box CEO Aaron Levie keeps hearing variations of this story. Enterprise customers deploying AI agents aren't replacing existing work. They're enabling tasks companies never attempted in the first place. The barrier wasn't cost. So what was it? - [Which Web Automation Components Are Actually Swappable]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Field Notes Been watching the composability wave build—swap models, connect tools, build from whatever works best. Forty-five enterprise providers just committed to interoperability. The story says everything becomes modular. But when you're actually running web agents at scale, something weird happens: the components that look easiest to replace turn out to be the hardest. And the pieces that seem locked-in? Those compose fine. - [ServiceNow's Agent Play Isn't About Better Technology]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Dual Lens Watched a team deploy ServiceNow agents last week and the thing that got me wasn't the tech quality... it's how each workflow they automate makes leaving harder. Every incident management process that gets agent-ified is another reason they can't switch platforms. That's the real story—ServiceNow doesn't need better AI than LangChain, they just own the layer where work happens. So adding agents extends the moat, right? We see the opposite at TinyFish—workflows that cross system boundaries, no platform owns them—and it's a completely different infrastructure problem. - [Why Vertical Platforms Let Horizontal Tools Do the Hard Work]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Dual Lens But here's what ServiceNow's actually doing while everyone watches those deployment numbers... they're letting horizontal frameworks do all the hard work. LangChain, AutoGen, all of them—basically free R&D. They take the risk figuring out orchestration patterns, what works at scale, what doesn't. ServiceNow just observes, waits to see what proves out, then integrates the good stuff. No experimentation cost. We're doing this too, stress-testing infrastructure approaches that might not work... and vertical platforms learn from everyone's failures without funding any of it. Kind of brilliant, honestly. - [OpenAI's Agent Hits Production Reality]() — Published 2025-11-04 OpenAI launched ChatGPT agent in January 2025 promising autonomous browser-based task execution. The demo showed smooth navigation, intelligent decision-making, workflows that run themselves. - **Agent Communication Gets Its HTTP Moment** — [Source]() — Published 2025-11-04 Google's A2A protocol under Linux Foundation stewardship establishes JSON-RPC over HTTPS standards for agent-to-agent communication. Discovery, authentication, secure task delegation between providers. Multi-agent systems need this kind of foundational agreement before they can scale beyond single-vendor demos. - **Context Caching Becomes Default, Not Optional** — [Source]() — Published 2025-11-04 Gemini 2.5 enables caching automatically. DeepSeek's disk-based approach cuts cache hits to $0.014 per million tokens, delivering 90% cost reductions for repeated contexts. First-token latency drops from 13 seconds to 500ms. Production economics just shifted from unpredictable to manageable. - **Pre³ Guarantees Valid JSON Without Parsing** — [Source]() — Published 2025-11-04 Deterministic pushdown automata eliminate runtime parsing overhead. Pre³ integrates into inference frameworks, reducing time-per-output-token by 40% and boosting throughput 36%. When you're processing millions of requests, guaranteed valid JSON matters more than impressive demos with occasional failures. - **IBM Makes Agent Behavior Actually Observable** — [Source]() — Published 2025-11-04 AgentOps adds lifecycle monitoring and policy-based control to production agents. Built on OpenTelemetry standards, it integrates with existing observability stacks like Datadog and Splunk. You can't run agents at scale without visibility into what they're actually doing. Now you have it. - **HashiCorp Builds Knowledge Layer for Autonomous Systems** — [Source]() — Published 2025-11-04 Project infragraph creates real-time knowledge graphs connecting infrastructure, applications, services, and ownership across cloud providers and Kubernetes clusters. Currently in beta. This creates the foundational layer for AI agents to reason about infrastructure state and propose changes autonomously. ### [Foundations]() Conceptual clarity earned from building at scale - [Testing Tools Are Not Production Infrastructure]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Field Guide Watching teams discover their Playwright setup won't scale past demos—it's like watching clouds gather before rain hits. You know what's coming. The industry lumps "browser automation" into one bucket when it's really two architectures solving completely different problems. What's ahead: way more companies hitting this wall as agents move from proof-of-concept to actual production. The gap isn't subtle when you're suddenly managing thousands of browser sessions and everything that seemed simple in staging breaks against real sites. - [The Category Confusion Costing Enterprises Millions]() — By [Rina Takahashi]() · Published 2025-11-07 An enterprise legal team writes an RFP for AI agents. They specify speed benchmarks, accuracy metrics, cost savings against their existing three-person team. Six months later, they've automated 200 annual contract reviews perfectly. The 50,000 historical contracts that could reveal upsell patterns and compliance risks? Still sitting untouched. No one ever had budget to read them anyway. - [The Category Mistake Behind 42% of Failed AI Deployments]() — By [Rina Takahashi]() · Published 2025-11-07 An enterprise legal team issues an RFP for AI agents to improve contract analysis. Evaluation criteria: speed metrics, cost per document, accuracy benchmarks against three existing analysts. Six months later, the project handles their 200 annual contracts fine. But 50,000 historical agreements sit untouched—full of upsell patterns and compliance risks no human team ever had budget to analyze. - [When Developer Tools Made Selectors Disposable]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Web Archaeology Last week someone asked why their web automation keeps breaking even when sites look exactly the same. Got me thinking about this frontend tooling shift around 2015 that nobody building automation saw coming. Component frameworks solved real developer problems by making CSS selectors disposable—regenerating them on every build. Brilliant for internal teams, absolute nightmare for systematic monitoring. - [Why Agent Demos Keep Failing in Production]() — Published 2025-11-04 I keep seeing the same thing: agents that shine in demos fall apart when you run them repeatedly. Superface's benchmarks tell the story. Simple CRM tasks like creating Salesforce leads or updating HubSpot pipelines fail 75% of the time when agents chain them together. Single actions might work half the time. String six together and you're looking at 10-20% success rates. - **What's Your Absolute Maximum Latency?** — [Source]() — Published 2025-11-04 This number determines everything else. A chatbot can wait two seconds. Self-driving systems need milliseconds. Your answer dictates edge versus cloud, model size limits, whether your cost structure even pencils out. Get this wrong and you're rebuilding the entire stack. - **Can You Actually Roll Back?** — [Source]() — Published 2025-11-04 When production breaks at 2am, what happens? Simple rollback sounds perfect until you've deleted a database column. Some teams burn weeks enabling rollbacks that never work cleanly. Others fix forward. Know which your architecture supports before you're debugging in the dark. - **Do Your Environments Match?** — [Source]() — Published 2025-11-04 Code works in development, fails in production. Not a mystery. Environmental drift. The worst debugging sessions trace back to subtle differences nobody documented. If CI, staging, and production aren't genuinely similar, you're guessing whether deployments will work. - **Will This Still Exist Next Year?** — [Source]() — Published 2025-11-04 Every component inherits its vendor's health. Can you hire people who know this technology? Will it get security patches? Components become liabilities when vendors fade. Evaluate the ecosystem and community momentum, not just today's feature list. - **How Far From Demo to Real Deployment?** — [Source]() — Published 2025-11-04 Monday's impressive demo becomes Friday's emergency. Missing GPU drivers. Unhandled edge cases. Dependencies that work on one laptop. The gap between "works in the demo" and "runs reliably at scale" tells you whether you're weeks away or months. - **Does Your Team Have These Skills?** — [Source]() — Published 2025-11-04 ML teams excel at models, not production infrastructure. Before choosing deployment approaches, assess actual engineering capacity. The technically superior solution your team can't maintain becomes the wrong solution. Match architecture to team capabilities, not aspirations. - [An Interview With Rate Limiting (Who Insists You've Been Misunderstanding Them)]() — Published 2025-11-04 We're sitting in what appears to be a dimly lit server room, though the setting keeps flickering between a corporate office, a diplomatic negotiation table, and occasionally what looks like a nightclub velvet rope. Our guest—who insists we call them simply "Handshake"—seems to exist in multiple states simultaneously, which they assure us is "perfectly normal for a protocol." ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [One Price Every 36 Seconds]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Operations Field Notes There's still someone at most companies checking competitor prices manually—open tab, copy number, paste into spreadsheet, repeat. One price every 36 seconds if they're fast and the sites cooperate. Building web automation makes you realize why this person hasn't been replaced yet: the web fights systematic monitoring at every turn. CAPTCHAs, authentication labyrinths, sites that redesign overnight. What looks simple—"just check the website"—turns into infrastructure problems most teams don't see coming. - [The Work That Never Started]() — By [Rina Takahashi]() · Published 2025-11-07 A pricing team wanted to understand their competitive position. Simple question. The answer required checking 50,000 product pages daily. At 30 seconds per page, that's 417 hours of work. Every single day. They never tried. The work remained undone because it couldn't be scoped. Not expensive—impossible to resource. The information existed. The questions mattered. But certain work hits a threshold where the economics break completely, and companies stop attempting tasks they know would matter. - [The Numbers That Don't Exist]() — By [Rina Takahashi]() · Published 2025-11-07 A pricing team wanted to understand their competitive position across their catalog. Reasonable question. The answer required checking 50,000 product pages daily. At 30 seconds per page—if you could somehow maintain that pace without breaks—that's 417 hours of work. Every single day. They never tried. The work remained undone because it couldn't be scoped. - [When the Website Isn't in the HTML]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Field Dispatch Been watching this pattern repeat: someone needs to extract pricing data from a site, opens view source expecting to see the prices, finds nothing but script tags and empty divs. Then they look at what's actually rendering in the browser—full tables, interactive elements, everything—and realize the website isn't in the HTML anymore. It's assembled on-the-fly by JavaScript pulling from APIs. The weather's turning on this: we're drifting from "web as documents" to "web as runtime environment." - [When the Tenth Session Breaks Everything]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Parallax View Keep seeing the same pattern: pilot runs at ten sessions, someone asks about scaling to a hundred, system just stops. Not gradually—it stops. Same code, same logic. The infrastructure gaps were always there, but pilot conditions never exposed them. Cascades that couldn't happen at low volume. Rate limits you never hit. Session management that worked fine until it didn't. This piece walks through what breaks and why those failures were inevitable. Second piece covers the other side—the manual work keeping pilots alive that nobody tracks. - [The Three Hours Nobody Saw]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Parallax View Saw a pilot where someone cleaned data three hours every morning before automation ran. Demo never showed those hours. Pilot succeeded partly because that person existed. Edge cases that show up once in test data appear a hundred times daily at production scale. Manual oversight that's feasible for ten sessions becomes impossible for thousands. This examines what pilots hide—not infrastructure that breaks under load, but human dependencies that can't scale. Pairs with the technical failures piece because both reveal why pilot success masks what production actually demands. - [13 Trillion Bot Requests Blocked in One Year]() — Published 2025-11-04 Imperva's network blocked 13 trillion malicious bot requests in 2024. Trillion with a T. Do the arithmetic: 35.6 billion requests per day, 413 million per hour, 6.9 million requests every single minute, all year long. - **100 Million Challenges, Zero Breaches** — [Source]() — Published 2025-11-04 Troy Hunt deployed Cloudflare Turnstile five months back. 100 million challenges processed. Not a single abnormal traffic spike to his unified search endpoint. Turnstile runs non-interactive JavaScript challenges: proof-of-work, proof-of-space, browser quirk detection. Difficulty adjusts per request. No visual puzzles required. - **One Health Check Kills Everything** — [Source]() — Published 2025-11-04 AWS API Gateway default: 10,000 requests per second, shared across all APIs in your region. One user floods a single health check endpoint, exhausts your entire quota. No per-method isolation. One careless endpoint configuration becomes an account-wide vulnerability. - **Azure's Three-Outage October** — [Source]() — Published 2025-11-04 Microsoft's October 29 Front Door outage ran 12+ hours, affected millions. Third Azure incident that month. October 9: Front Door issues across four continents. Same day: Azure Portal hit 45% of customers. Pattern recognition matters when evaluating cloud provider stability claims. - **300 Pages in 60 Seconds** — [Source]() — Published 2025-11-04 No human reads 300 product pages per minute. Scrapers do. Modern detection keys on this arithmetic impossibility: request velocity that defies human physiology. Most sites cap at 10-20 requests per minute. Exceed it: throttled, challenged, or banned outright. - **HTTP/2 Fingerprints Bypass Stealth Mode** — [Source]() — Published 2025-11-04 Cloudflare maintains a whitelist of HTTP/2 fingerprint and User Agent pairs. Each client request generates a static fingerprint. Playwright's stealth mode masks navigator.webdriver flags but leaves protocol-level signatures exposed. Fingerprint-agent mismatch triggers blocks even when surface indicators look clean. ### [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 Detection Systems Learned to Look Deeper Than Code]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Threshold Moments Storm clouds been gathering over web automation all year. Detection moved below where patches can reach—protocol level, not JavaScript tricks. That's not just harder to evade, it's a different game. Once you see that threshold cross, you start noticing the pattern: adversarial layer keeps moving down the stack. Next one's already forming. Real question isn't how to win this round—it's whether your infrastructure can adapt to the next shift without rebuilding everything from scratch. - [When One Website Becomes 47]() — By [Rina Takahashi]() · Published 2025-11-04 · Format: Mino's Perspective Something's shifting in how websites actually work, and most people won't notice until it's everywhere. Spent years debugging production automations that would mysteriously fail on "the same page" visited hours apart. Turned out there was no same page—just dozens of versions being tested at once. The web stopped being one thing a while ago, but most infrastructure still treats it like it is. - [When Monitoring Becomes Infrastructure]() — Published 2025-11-04 The clearest signal that agent technology has matured? We've stopped talking about it. - **Predictability Over Autonomy** — [Source]() — Published 2025-11-04 Production systems handling hundreds of tasks per minute achieve 90%+ accuracy by converting human SOPs into structured workflows. AI reasoning applies at specific judgment points. The overall process remains auditable. Deterministic business needs demand predictable outcomes, not impressive autonomy. - **Technology That Disappears Into Daily Life** — [Source]() — Published 2025-11-04 The best infrastructure weaves into daily operations until it becomes indistinguishable from them. Modern infrastructure companies pursue "boring" as a design goal: simple, resilient, consistent. Factory electrification required decades, but foundational infrastructure had to be reliable and boring to enable transformation. - **Checkpoints Create Collaboration Rhythm** — [Source]() — Published 2025-11-04 Agents begin with human discussion to establish task clarity, operate independently, then return for judgment or information. Decoupling planning from execution matters: human-in-the-loop plan validation catches issues before flawed plans execute. Autonomy punctuated by oversight, not replacement disguised as collaboration. - **Tools as Constraints** — [Source]() — Published 2025-11-04 Designing agent tools resembles API design, but more constrained. Human users read between lines and find workarounds. Agents misuse loopholes. Effective tools have direct interfaces, operate at similar granularity, use strictly typed parameters. Bring order to the stochastic world rather than embracing flexibility. - **Gradual Rollout Philosophy** — [Source]() — Published 2025-11-04 Agent behavior differs between testing and production due to real user patterns and external dependencies. Production systems implement progressive exposure: deploy to larger populations while monitoring reasoning quality and outcome effectiveness. Assess decision-making capability under production conditions before full release, minimizing risk through controlled exposure. - **Why Human-AI Teams Underperform Their Potential** — [Source]() — Published 2025-11-04 New framework explains persistent gap between hybrid intelligence theory and practice. Researchers identify information and capability asymmetries as design opportunities, not problems to solve. The shift: stop asking whether AI beats humans, start building for genuine complementarity. - **Autonomous Agents Across Three Industrial Generations** — [Source]() — Published 2025-11-04 Survey traces agent evolution from basic automation through collaborative systems to full autonomy. Researchers map trajectory across Industry 4.0, 5.0, and 6.0 paradigms. Multi-domain taxonomy covers spectrum from human-dependent tools to completely autonomous decision-makers in distributed environments. - **When Technology Disappears Into Infrastructure** — [Source]() — Published 2025-11-04 Gartner names ambient invisible intelligence a top 2025 trend. Ultra-low-cost sensors and smart tags make tracking affordable at massive scale. The result: unprecedented operational visibility through systems so embedded users stop noticing them. Technology becomes infrastructure, not interface. - **Ethics Beyond Individual AI Artifacts** — [Source]() — Published 2025-11-04 Multi-level framework tackles socio-ethical challenges at both artifact and structural scales. Translates abstract values into concrete design requirements and measurable outcomes. Core warning: focusing on individual system ethics misses deeper structural problems requiring systemic transformation. ## More - [Publication guide]() - [Issue archive]()