--- title: "Issue 19" canonical: "https://current.tinyfish.ai/issue/19" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/19" issue: 19 description: "Your agents are succeeding brilliantly. That might be the problem." authors: ["Sable Whitford","Nora Kaplan","Rina Takahashi"] published: 2026-03-08 updated: 2026-11-05 articles: 90 --- # 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 19 Your agents are succeeding brilliantly. That might be the problem. ### [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. - [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. - [Faria Huq and the Reason Behind a Human Step]() — Published 2026-10-30 When a person takes control from an AI agent driving their web browser, the system records a human step, but not why. Faria Huq and her Carnegie Mellon collaborators have built tools around these moments: an agent that can't find the right forum, a shopper who realized the brand mattered only after seeing the agent's pick. Their system learns to predict when someone will step in. The reason it sets aside would tell a team whether to fix the agent, the website, or its understanding of the person. - [Lebanon's Two Clocks and the Question Calendars Don't Ask]() — Published 2026-10-30 On March 26, 2023, Lebanon woke up on two clocks. The caretaker government had postponed daylight saving three days earlier. Churches and TV channels moved forward anyway, government offices stayed put, and most phones jumped on their own. Middle East Airlines printed every departure an hour earlier, so its planes still left at the same moment. That week made each calendar show what it had been storing: an exact instant, or a time on the wall that politics can move. Software assistants now make that choice without asking anyone. ### [Market Pulse]() Reading the agent ecosystem through a practitioner's lens - [The Browser Was Never Neutral. Now Something Is Testing the Terms.]() — By [Nora Kaplan]() · Published 2026-03-04 · Format: The Control Layer On August 19, 2025, Amazon blocked Perplexity's browser from its marketplace. Within twenty-four hours, Perplexity shipped a workaround. Amazon blocked again. By November, they were in federal court. That escalation happened fast enough to tell you something about the stakes. - **Perplexity Comet Brings Agentic Browsing to iOS** — [Source]() — Published 2026-03-04 Chromium-based AI browser, free worldwide since October 2025. iOS launch scheduled March 11, 2026. Runs Claude Sonnet 4.6 for Pro subscribers, Opus 4.6 for Max. Handles full browsing workflows autonomously, turning web sessions into conversational interactions where the agent clicks, scrolls, and transacts on the user's behalf. - **Dia Gets Enterprise Backing via Atlassian Acquisition** — [Source]() — Published 2026-03-04 The Browser Company shelved its power-user Arc browser to build Dia, a Chromium-based AI browser aimed at everyday workflows. Atlassian acquired the company for a reported $610M, folding Dia into its knowledge-worker ecosystem with planned enterprise integrations. - **Opera Neon Runs Agents Locally, Eyes Europe** — [Source]() — Published 2026-03-04 Publicly traded Opera opened Neon access in December 2025. Its "Neon Do" function operates within user sessions locally rather than routing through the cloud. A February 2026 update added automatic agent selection. European server infrastructure gives it a pitch to GDPR-conscious markets. - **OpenAI Building a Browser Around Operator Integration** — [Source]() — Published 2026-03-04 OpenAI is developing a Chromium-based browser to embed its Operator web agent directly into browsing, per Reuters. Built by former Google Chrome engineers. Referred to publicly as ChatGPT Atlas. With 500 million weekly ChatGPT users, the potential install base is already enormous. - **Chrome and Safari Still Own the Traffic** — [Source]() — Published 2026-03-04 Google and Apple together hold roughly 83–85% of global browser traffic, per StatCounter. Google shipped Chrome Auto Browse in January 2026, adding agentic capability inside the dominant browser rather than yielding the layer. Apple has made no comparable announcement. - **Amazon Sues Perplexity Over Comet's Shopping Automation** — [Source]() — Published 2026-03-04 Filed November 4, 2025 in San Francisco federal court. Amazon alleges Comet disguises itself as Chrome, makes purchases without disclosing bot activity, and evaded Amazon's technical barriers. Perplexity argues user agents carry the user's own permissions. The outcome could set platform access rules for autonomous agents broadly. - **Principal-Agent Dynamics and Digital Platform Economics in the Age of Agentic AI** — [Source]() — Published 2026-03-04 Uses principal-agent theory to map how autonomous agents destabilize two-sided platform economics. When agents act on behalf of users, they bypass ads, recommendations, and the attention-capture mechanisms platforms rely on to monetize both sides. - **Amazon.com Services v. Perplexity AI — Federal Complaint (N.D. Cal., November 2025)** — [Source]() — Published 2026-03-04 The first federal lawsuit squarely targeting a browser agent's autonomous behavior on a third-party platform. Amazon alleges Perplexity's agent concealed its identity and evaded technical barriers to transact on users' behalf, framing this as computer fraud and contract violation. - **Fine-Tuning Aligned Language Models Compromises Safety, Even When You Don't Intend It To** — [Source]() — Published 2026-03-04 Princeton and Stanford researchers showed that fine-tuning reliably degrades safety alignment, even with benign data. Adversarial fine-tuning on just 10 examples bypassed guardrails at under $0.20. Peer-reviewed at ICLR 2024, with findings replicated independently across model families. - **WebVoyager Browser Agent Benchmark Leaderboard (February 2026)** — [Source]() — Published 2026-03-04 The standard browser agent benchmark now shows top performers completing 87–97% of web tasks across 643 scenarios. But benchmark conditions exclude bot detection, CAPTCHAs, and fingerprinting, so production success rates on defended sites are materially lower. - [Amazon's Lawsuit Against Perplexity Frames Agent Browsing as Computer Fraud]() — Published 2026-03-04 "Undisclosed bot activity that masquerades as human clicks." That phrase, buried in Amazon's November 2025 complaint, deserves more attention than the lawsuit itself. Amazon alleges Perplexity's Comet browser spoofed standard Chrome user-agent strings to pass automated browsing off as human activity, then updated its evasion techniques within 24 hours of new detection measures going live. - **Comet: Perplexity's AI Browser Gets Personal** — [Source]() — Published 2026-03-04 Grounds the agentic browser concept in actual product decisions, showing where Comet bets on personalization and where Chrome's ecosystem gravity still wins. - **The State of AI & Browser Automation in 2026** — [Source]() — Published 2026-03-04 A practitioner survey of how browser automation has quietly matured into infrastructure, with useful detail on the reliability gaps that still separate demos from production. - **Amazon Sends Legal Threats to Perplexity Over Agentic Browsing** — [Source]() — Published 2026-03-04 The first confirmed legal confrontation over agentic browsing, useful for understanding how platform owners are drawing lines around automated access. - **When a Cease-and-Desist Letter Calls Your AI a 'Computer Fraud'** — [Source]() — Published 2026-03-04 A legal practitioner walks through how existing ToS clauses become liability instruments against web agents, with implications for anyone building in this space. - **A New Browser War for the Agentic Web** — [Source]() — Published 2026-03-04 Applies a competition-policy lens to explain why agents threaten the attention economy and why incumbent leveraging by Google and Apple is structurally predictable. - **Amid New Competition, Chrome Speeds Up Its Release Schedule** — [Source]() — Published 2026-03-04 The freshest signal that Google views agentic browsers as a competitive threat worth responding to, not just a curiosity worth watching. ### [Foundations]() Conceptual clarity earned from building at scale - [What "Platform Shift" Tells You to Build]() — By [Nora Kaplan]() · Published 2026-03-04 · Format: The Metaphor Problem At MWC this week, Qualcomm's CEO described an agent-centric future. Lenovo's EVP compressed the idea to three words: "apps to intent." Neither executive used the phrase "platform shift." Within hours, analyst notes had supplied it for them. - [The Three Infrastructure Problems Hiding Inside "Agent-Ready"]() — By [Rina Takahashi]() · Published 2026-03-04 · Format: What Ready Requires MCP has 97 million monthly SDK downloads. ChatGPT, Gemini, and Copilot have all adopted it. The connective plumbing for agents is getting standardized at a pace that suggests the hard part is over. Meanwhile, nearly half of organizations still can't make their own data searchable by the systems that plumbing is supposed to connect. - **Qualcomm's CEO Declares 2026 the Year of the Agent** — [Source]() — Published 2026-03-04 Worth reading for the specific language about agents replacing apps. The framing tells you where the narrative is headed. - **Deloitte on Why Layering Agents onto Old Workflows Breaks Faster** — [Source]() — Published 2026-03-04 Only 11% of organizations have agents in production. Deloitte's argument for why: value requires redesigning operations, and most teams skipped that part. - **How the Model Context Protocol Became Infrastructure** — [Source]() — Published 2026-03-04 The adoption story beneath the hype. MCP won by growing up in public, and the trajectory matters more than the current spec. - **MCP Lands on the Executive Agenda, Security-First** — [Source]() — Published 2026-03-04 Fewer than 4% of RSA Conference MCP submissions focused on opportunity. The security community is concentrating almost entirely on exposure. - **Seven Ways Multi-Agent Systems Break in Production** — [Source]() — Published 2026-03-04 A vendor blog, so calibrate accordingly. But the failure taxonomy here (context overflow, chained reliability decay, race conditions) reflects real production experience worth absorbing. - **The Capital Thesis: $52 Billion by 2030** — [Source]() — Published 2026-03-04 MarketsandMarkets sizes the AI agents market at $7.84B in 2025, growing to $52.62B by 2030. That's a 46.3% CAGR, and it sits at the conservative end of analyst estimates. The money is betting loudly. Production teams are the ones who have to cash the check. - **Gartner Sees Eightfold Agent Jump by 2026** — [Source]() — Published 2026-03-04 From under 5% in 2025 to 40% of enterprise apps featuring task-specific agents by end of 2026. An eightfold jump in twelve months. Worth noting: Gartner itself warns against "agentwashing," where basic assistants get relabeled as agents. The same firm issuing the forecast is hedging the definition. - **Majority of Organizations Now Run Production Agents** — [Source]() — Published 2026-03-04 LangChain's 2026 State of AI Agents report puts 57% of organizations running agents in production, up from 51% last year. Large enterprises lead at 67%. Quality remains the top deployment barrier, cited by 32% of respondents. The momentum is genuine. So is the struggle to keep things working well. - **Four in Ten Agent Projects Won't Survive** — [Source]() — Published 2026-03-04 Gartner predicts over 40% of agentic AI projects will be canceled by end of 2027. The culprits: costs that blow past estimates, business value that stays fuzzy after deployment, and risk controls too thin to catch failures before they compound. Shipping is one thing. Sustaining is another. - **Half of Organizations Say Data Isn't Agent-Ready** — [Source]() — Published 2026-03-04 Deloitte's 2025 survey found 48% of organizations cite data searchability and 47% cite reusability as blockers to AI automation. Enterprise data architectures were built around ETL pipelines and warehouses. Agents need business context to act, and most data stacks weren't designed to provide it on demand. - **One in Five Has Mature Agent Governance** — [Source]() — Published 2026-03-04 Deloitte's 2026 State of AI in the Enterprise finds only 20% of companies have a mature governance model for autonomous agents. The other 80% are assembling oversight frameworks while their agents are already live. Governance is trailing deployment by a widening margin, and the agents aren't waiting around. - **MCP Crosses 97 Million Monthly SDK Downloads** — [Source]() — Published 2026-03-04 Over 10,000 active servers now speak MCP, with native client support in ChatGPT, Claude, Gemini, and Copilot. Anthropic, OpenAI, Google, and Microsoft all back the protocol through the Agentic AI Foundation. The connective tissue of the agent stack is no longer theoretical. - **Microsoft Folds AutoGen Into a Unified Framework** — [Source]() — Published 2026-03-04 AutoGen is now in maintenance mode. Microsoft merged it with Semantic Kernel into Agent Framework, with the Release Candidate shipping February 2026 and GA targeted this quarter. Meanwhile, the broader landscape is sorting itself: LangGraph owns stateful orchestration, CrewAI targets fast multi-agent prototyping. Distinct lanes, finally. - **Browser Use Hits 89.1% on WebVoyager Tasks** — [Source]() — Published 2026-03-04 On WebVoyager's filtered 586-task subset, Browser Use leads open-source browser agents at 89.1% success. Playwright holds 45.1% adoption among QA professionals. The interesting pattern: AI reasoning layers are landing on top of traditional automation, pushing selector maintenance from 15–25% of effort down below 5%. - **First Community Census Counts 104,504 Registered Agents** — [Source]() — Published 2026-03-04 The Universal Agent Registry's inaugural monthly report indexes 104,504 agents across 15 registries. Fetch.ai's AgentVerse holds the largest share at 34.8%. Text generation still dominates with 61,649 entries, but 19 distinct capability categories suggest the ecosystem is growing past its chatbot origins. - **MCP Moves to Linux Foundation's Agentic AI Foundation** — [Source]() — Published 2026-03-04 Anthropic donated MCP to the Agentic AI Foundation, a Linux Foundation directed fund co-founded with Block and OpenAI. Google, Microsoft, AWS, Cloudflare, and Bloomberg signed on as backers. Salesforce and Microsoft Dynamics 365 have already added native MCP support. The governance question is now distributed across 50+ enterprise partners. - **Four Core Metrics Emerge for Agent Observability** — [Source]() — Published 2026-03-04 Practitioner consensus is coalescing around success rate per task, average steps to resolution, tool call accuracy, and cost-to-value ratio of reasoning loops. Stanford CFRM research suggests structured evaluation can reduce production failures by up to 60%. The hard part: benchmark scores and production reliability still tell very different stories. ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [Every Metric Is Green. That's the Problem.]() — By [Rina Takahashi]() · Published 2026-03-04 · Format: The Wrong Success An autonomous customer-service agent started approving refunds it shouldn't have. Review scores climbed. Satisfaction metrics improved. Every dashboard confirmed the system was performing beautifully, and it kept performing beautifully while giving away money, thousands of times per hour. Nobody pulled the kill switch. The monitoring stack couldn't distinguish what was happening from success. Agents turned Goodhart's Law into an operational crisis. - [What Browser Use Found When They Stopped Looking at Screenshots]() — By [Rina Takahashi]() · Published 2026-03-04 · Format: Builder Profile Most browser agent frameworks begin by taking a screenshot. Feed pixels to a model, ask it where to click. Magnus Müller and Gregor Žunič, two ETH Zurich engineers who built a prototype in four days and watched it collect 50,000 GitHub stars, skipped the screenshot and read the page's structure as text. - [The Whiteboard Is Losing — A Conversation with the Person Translating Business Intent into Agent Objectives]() — Published 2026-03-04 *Gil Trotter does not exist, but his job does. Someone at your company is almost certainly doing it right now, probably without the title, possibly without knowing it. What follows is a conversation with a fictional practitioner built from real role descriptions, documented failure modes, and the operational reality surfacing across organizations deploying customer-facing agents. The specification gap he describes is not invented. The scenarios are.* - [75+ Observability Startups Are Racing to Build Monitoring. Most Are Measuring the Wrong Thing.]() — Published 2026-03-04 Cleanlab's 2025 production survey puts the number at 5%. That's the share of AI agents in production with what anyone would call mature monitoring. Meanwhile, CB Insights ranks agent observability as the most commercially dynamic category in generative AI, with 75-plus startups competing for the space. So capital is flooding in. The question worth sitting with: flooding in toward what, exactly? - **Silent Failure at Scale: The AI Risk That Can Tip the Business World into Disorder** — [Source]() — Published 2026-03-04 The documented production failures here expose what monitoring infrastructure was designed to catch, and what it structurally cannot. - **Human-in-the-Loop Has Hit the Wall. It's Time for AI to Oversee AI** — [Source]() — Published 2026-03-04 One enterprise modeled needing four times its current review headcount just to keep up with approvals. The economics broke before governance could scale. - **AI Agents Are Transforming Enterprise Operations and Driving Infrastructure Demand** — [Source]() — Published 2026-03-04 The 80/40 split — strategic priority vs. unmeasurable ROI — goes a long way toward explaining why observability now leads generative AI in deal activity. - **AI Just Leveled Up and There Are No Guardrails Anymore** — [Source]() — Published 2026-03-04 Anthropic replaced binding safety commitments with "nonbinding targets" the same month agents leapt from chatbot to executive assistant. The timing says more than either story alone. - **The State of AI and Browser Automation in 2026** — [Source]() — Published 2026-03-04 Authentication walls, bot detection, and layout shifts produce failure modes that look nothing like model drift. Worth reading for the infrastructure-layer perspective the other pieces don't reach. ### [Echoes]() Past infrastructure decisions echoing in today's production realities - [The Decimal Point That Won't Move]() — By [Rina Takahashi]() · Published 2026-03-04 · Format: Virtues Into Chains In most modern programming languages, 0.1 plus 0.2 does not equal 0.3. A rounding artifact, harmless in most contexts, catastrophic when you're settling millions of bank transactions to the penny. In 1959, a government-backed committee made a handful of specific design choices that solved this problem completely. Those choices now sit underneath trillions of dollars in daily commerce and 95% of ATM transactions. Organizations spend years and hundreds of millions trying to move off them. About 70% of the time, the migration fails. The code works exactly as designed. - [The Replacement That Never Came: How Decades of "Almost Done" Left COBOL's Knowledge Walking Out the Door]() — Published 2026-03-04 Every decade since the 1980s, someone declared COBOL nearly dead. Universities dropped it from curricula. Training pipelines dried up. And the systems kept running, processing most of the world's credit card transactions on logic nobody was teaching anyone to read. - [My Best Work Is When Nothing Happens — A Conversation with the Person Behind $3 Trillion in Nightly Batch Jobs]() — Published 2026-03-04 Every weeknight, somewhere around 11 p.m. Eastern, a batch job kicks off at a bank you've almost certainly used. It reconciles the day's transactions, settles accounts, flags exceptions, and generates the reports that let the next business day begin. The code doing this was largely written before the programmers running it were born. It still observes line lengths dictated by the physical dimensions of a punch card that stopped being manufactured in the 1980s. And it processes roughly $3 trillion in daily commerce. - [The Fix That Lasted Forever]() — By [Nora Kaplan]() · Published 2026-03-04 · Format: Patch, Don't Replace Someone spotted the problem in 1958. Nobody acted for forty years. When Y2K finally forced the largest coordinated software effort in history, organizations faced a real choice: replace their aging COBOL systems or patch them. The money was there. The motivation was existential. But the deeper organizations looked inside their own code, the less replacement looked like the rational bet. The fix they chose is still running. - [Better Tools for the Same Trap]() — By [Nora Kaplan]() · Published 2026-03-04 · Format: Patch, Don't Replace AI coding agents are the most capable tools ever aimed at COBOL modernization. They can read legacy code, document undocumented logic, translate between languages. Early results are promising. But every previous modernization wave arrived with genuinely capable tools, too, and roughly 70% of legacy migration projects still fail. The complexity that defeats them extends well past the source code. - **The Mainframe Moment: How AI-Driven Modernization Is Reshaping the COBOL Economy** — [Source]() — Published 2026-03-04 Goes past the technical debate to map the billion-dollar economy that formed around not replacing COBOL. The incentive structures alone are worth your time. - **What Is COBOL Modernization?** — [Source]() — Published 2026-03-04 IBM's explainer doubles as a defense of the complexity it sells. Read it as a primary source on why the incumbent frames modernization as a decades-long journey, not a migration. - **The COBOL Crisis** — [Source]() — Published 2026-03-04 The clearest account of two clocks running simultaneously: the talent pool shrinking and the codebase calcifying. Useful for understanding why urgency keeps rising without producing action. - **Mainframe Careers Are Changing, Not Disappearing: What to Expect** — [Source]() — Published 2026-03-04 Reframes the "talent shortage" as a translation problem. The scarcest skill isn't writing COBOL; it's bridging the gap between mainframe logic and cloud-native thinking. - **COBOL Modernisation Just Got an AI Shortcut — and the Market Noticed** — [Source]() — Published 2026-03-04 A good reconstruction of how a single AI demo wiped billions off IBM's market cap. The speed of the market reaction tells you something about how thin the confidence layer around mainframe economics really is. - **Lost in Translation: What the AI Code Debate Keeps Getting Wrong** — [Source]() — Published 2026-03-04 IBM's counterargument that translating code isn't the same as modernizing a platform. More substantive than you'd expect from a company playing defense. Pair it with the AI News piece above. ### [Vision]() Where human-AI collaboration is heading - [The Rise of Ghost Institutions]() — By [Nora Kaplan]() · Published 2026-03-04 · Format: Acquired Taste Somewhere right now, a contract worker is choosing between two AI-generated paragraphs. The rubric says "quality." It says "helpfulness." Nothing about the task registers as a cultural act. In some cases, fewer than a hundred people like this shape how a language model behaves. Meanwhile, AI agents facilitated $22 billion in sales last Black Friday alone. - [What Counts as Good When Agents Decide]() — By [Rina Takahashi]() · Published 2026-03-04 · Format: Default Settings AI shopping agents describe their recommendations as "high-quality" and "unbiased." Neither term is defined. At small scale, that's a design choice. At millions of daily purchase decisions, it starts functioning like a standard. One that nobody published, nobody voted on, and nobody can fully inspect. Products meeting the criteria get recommended. Products that don't quietly stop appearing. The mechanism by which filtering criteria harden into definitions of quality is worth tracing carefully. - [What Stops Being Found When Every Agent Agrees]() — By [Rina Takahashi]() · Published 2026-03-04 · Format: Default Settings If different AI agents applied genuinely independent judgment, the filtering described in our companion piece would be uncomfortable but survivable. A product missed by one agent might surface through another. Recent research suggests otherwise: major language models converge not just in what they get right but in what they get wrong, correlating more with each other than with ground truth. The long tail of the web survived on human serendipity and dead-end browsing. Agents don't wander. - [The Ceramicist Below the Fold]() — Published 2026-03-04 *Delia Fontaine-Park does not exist. Neither does Sunday Object, the mid-market DTC kitchen brand where she supposedly runs product. But the problem she describes is real, documented, and currently unfolding across thousands of brands that haven't noticed it yet. We built her from interviews, trade reporting, and emerging research on agent-driven commerce to explore a question real operators are only beginning to ask: what happens when your product is fine, your reviews are strong, your customers are loyal, and you've simply stopped existing in the places where new customers now look?* - [The Supply Chain of Agent Taste]() — Published 2026-03-04 When an agent recommends a hotel or a vendor, that recommendation arrives carrying the accumulated preferences of people who never thought of themselves as tastemakers. - **The AI Agent Identity Crisis: 80% of Agents Can't Prove Who They Are, and 80% of Sites Don't Check** — [Source]() — Published 2026-03-04 Agent taste is meaningless if merchants can't tell a legitimate recommender from a spoofed imposter at scale. - **The $236 Billion Trust Gap: Why the Agent Economy Needs Its Own KYC Moment** — [Source]() — Published 2026-03-04 WEF argues agent commerce stalls without identity frameworks. Preference only matters once provenance is solved. - **AI Bot Traffic Surges 300% as Akamai and Visa Race to Build Commerce Trust Rails** — [Source]() — Published 2026-03-04 Agent traffic is already reshaping commerce patterns faster than anyone can build the trust layer underneath. - **Forrester: 30% of Enterprise Vendors Will Ship MCP Servers, Redesigning Software for Agent-First Access** — [Source]() — Published 2026-03-04 What agents can see determines what they recommend. MCP servers are quietly becoming the new shelf placement. - **From Discovery to Influence: How Brands Are Learning to Optimize for Agent Preference** — [Source]() — Published 2026-03-04 The supply side of agent taste is here: structured data and third-party validation now govern what agents surface. ## More - [Publication guide]() - [Issue archive]()