--- title: "Issue 54" canonical: "https://current.tinyfish.ai/issue/54" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/54" issue: 54 description: "One grocery-store problem followed us through this whole issue. When an inventory system believes a shelf holds stock that isn't there, it never places a reorder, and that reorder was the one event that would have exposed the empty shelf. The mistake quietly switches off its own alarm. Every agent you put into production reads some record and acts on it. So go look at your checks and ask an uncomfortable question: which of them only fire on something the failure itself prevents?" authors: ["Sable Whitford","Sevda Polat"] published: 2026-11-09 updated: 2026-11-14 articles: 109 --- # 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 54 One grocery-store problem followed us through this whole issue. When an inventory system believes a shelf holds stock that isn't there, it never places a reorder, and that reorder was the one event that would have exposed the empty shelf. The mistake quietly switches off its own alarm. Every agent you put into production reads some record and acts on it. So go look at your checks and ask an uncomfortable question: which of them only fire on something the failure itself prevents? ### [Daily Brief]() Your daily digest of web developments through the lens of enterprise agent deployment - [Your ChatGPT Agent Has a Credit Card Now. United and Amazon Would Like a Word.]() — By [Sable Whitford]() · Published 2026-11-14 OpenAI's **ChatGPT Wallet is live for US Pro subscribers**, and it lets the agents OpenAI calls "dots" both pick a product and pay for it. You store a US Visa, Mastercard or Amex once. At checkout you see the merchant, items, total and card, then approve with Face ID or a fingerprint. The merchant gets a payment token instead of your card number. On Wednesday, Veterans Day, United said it will block agents from booking tickets. Amazon already bars them. OpenAI's response is a Commerce Partner Playbook for merchants. My read: the stores blocking agents can't see agent traffic yet, and buyers approve a total without seeing why the agent chose it. - **llama.cpp hits 1.0, and versions mean something** — [Source]() — Published 2026-11-14 llama.cpp and ggml, the engines behind a huge share of run-it-yourself AI, shipped v1.0.0 yesterday after three and a half years. They now use strict semver, so a major version bump means the C API broke. Thirty-five deprecated functions are gone, FP8 support grew, and backends from CUDA to WebGPU got fixes. SGLang announced its own 1.0 too, so local inference got two birthdays on the same day. - **A manifesto against AI slop just dropped** — [Source]() — Published 2026-11-14 Speakers and attendees at AIE Code Summit 2026 published a No AI Slop manifesto yesterday, modeled on 2001's Agile Manifesto. Their test: if you spent less effort making something than your reader has to spend decoding it, it's slop. Half the replies are practical fights about reviewing AI-written PRs. The other half doubt a manifesto has ever beaten a quarterly target. - **DeepSeek breaks its app into Lego bricks** — [Source]() — Published 2026-11-14 Yesterday DeepSeek started selling the parts inside its own app as standalone Search, Fetch, OCR, speech-to-text and text-to-speech APIs. It also launched three endpoints running deepseek-flash (V4.1-Flash-1017) and released an open-source frontend demo. Developers get the same bricks DeepSeek built with, and they're free to build something completely different. - **Nvidia: 607.8 tokens/sec on two DGX Stations** — [Source]() — Published 2026-11-14 Nvidia has a write-up for when one GPU isn't enough. Its answer is to serve GLM-5.3 across two DGX Stations, its deskside AI workstations, using vLLM. The headline number is 607.8 tokens per second. That's serious multi-GPU serving on hardware that fits beside your chair. - **Google Research puts Antigravity agent teams to work** — [Source]() — Published 2026-11-14 Google Research posted about using /teamwork, a feature in Google's agent-first coding environment Antigravity, to push science and engineering work forward. Agent teams are the pitch of the season. My question for every team of agents stays the same: which human reads the output before it ships? - **Terminactory: an idle factory in your terminal** — [Source]() — Published 2026-11-14 Someone built an idle factory game that runs entirely in the terminal, written in Rust with ratatui. The numbers go up in the same window where your logs scroll, so from across the room it looks like work. It's Saturday, so your weekend is already gone, and Monday's standup attention may go with it. - **Ground stop update: everyone asked about the AI** — [Source]() — Published 2026-11-14 New details on yesterday's stop: it began at 3:54 p.m., lasted about four hours, and was only the third nationwide ground stop in FAA history. Secretary Sean Duffy cited a communication issue affecting all Air Route Traffic Control Centers. On Tuesday the FAA had expanded its $875M AI-supported SMART tool to all 22 centers, and DHS Secretary Markwayne Mullin said officials were checking for a link. Vendor Air Space Intelligence says SMART wasn't the cause, and Duffy says it "isn't plugged in." The cause is officially under investigation. Until there's a postmortem, the usual ops reflex applies, and whatever shipped Tuesday is the first suspect. - **Microsoft 365 spent a day as Microsoft 364** — [Source]() — Published 2026-11-14 Outlook went down Thursday afternoon, and Microsoft was still restoring it nearly a day later. SharePoint and Copilot were largely out through yesterday's US morning. Microsoft apologized yesterday and says today everything is fully restored. It blamed an issue from an earlier network capacity upgrade, "not related to cyber" or AI agents. People roasted the status updates, and a few inboxes were quietly hoping for 363. - **Claude stumbled too, apparently by association** — [Source]() — Published 2026-11-14 Claude had a rough Friday as well, with high error rates across claude.ai, Claude Code and the developer console. Anthropic's status page reportedly blamed a Microsoft 365 integration problem plus authentication trouble, so it appears linked to the Outlook mess. If a vendor sits in your login path, their outage ends up on your status page. - **React Server Components RCE now has a CVE** — [Source]() — Published 2026-11-14 The remote code execution bug in React Server Components got its CVE Thursday: CVE-2026-126187, rated Critical at 9.5 on CVSS 4.0. The fixed versions are 19.0.9, 19.1.10, 19.2.9 and 19.3.1, and npm audit and Dependabot are flagging it. Git 2.56.1 also patches six flaws. Upgrade before your weekend gets interesting. - **EFF: your privacy now depends on strangers' gadgets** — [Source]() — Published 2026-11-14 The EFF says tech companies are trying to redefine privacy again. It points to Apple's Audio Intelligence turning every Apple Watch into a transcription machine, Amazon's Bee devices that record and transcribe everything, and dozens of always-recording pendants and rings. Your privacy now depends on everyone around you knowing how each of those devices signals that it's recording. The EFF also says to be skeptical of "inevitable," since plenty of these gadgets turn out to be vaporware. - **Apple HomeView: great recipes, sometimes-frozen Siri** — [Source]() — Published 2026-11-14 Apple's first smart home display went on sale yesterday. It's a 7-inch screen on a speaker base that handles music, FaceTime and smart home control, all run by Siri AI. Reviewers love the recipe conversion, find the app lineup thin, and caught Siri freezing now and then. The verdict is not to expect a tablet. It's also one more microphone in your kitchen. - **Nvidia slides as Huang's exit unsettles AI funding** — [Source]() — Published 2026-11-14 Nvidia shares fell 2% yesterday after dropping more than 4% Thursday, following Jensen Huang's resignation as CEO. VCs and founders worry his departure leaves a real gap in funding for AI startups. Hugging Face's CEO vowed to keep fighting for open source AI. The industry is now watching who backs the next round of AI companies. - **LangChain reportedly lands $300M at $3B valuation** — [Source]() — Published 2026-11-14 LangChain has reportedly raised a $300M Series C at a $3B valuation, led by CapitalG. Its LangSmith platform reportedly serves 7,000 customers. Money keeps flowing to agent tooling, and tracing is a big reason why: once agents act in production, teams will pay to see what those agents actually did. - [Revenue Is Number 28, and Someone Has to Tell the SQL Agent]() — Published 2026-11-05 Ask an AI agent a question about your company's data, and it writes a database query, runs it, and returns a number. New research tested how often that number is right. Running was rarely the problem: every system tested wrote queries that ran at least 99% of the time. Yet apart from the researchers' own system, none answered even half the questions correctly. The wrong answers look as exact as the right ones, and nothing warns you which is which. This close read asks what a query that runs actually tells you. - [When AI Models Get Cheaper, Who Keeps the Savings?]() — Published 2026-11-05 When a customer pays SAP a fixed price for each task SAP's AI handles, the model doing the work can change while the price stays put. SAP is making that swap, moving work onto cheaper models and keeping the difference. In a written answer to Reuters, it said so plainly: "The goal is healthy margins as scale grows, not an absolute reduction in AI expenditure." - [Why QUIC Shipped a Version 2 With No New Features]() — Published 2026-11-05 Internet protocols were built to be upgradable. In 2017, a Chrome update tested that in Montgomery County, Maryland, where upwards of 30 percent of the district's roughly 50,000 school Chromebooks got stuck looping at the login screen, locking students out. - [Phantom Stock and the Checks That Never Fire]() — Published 2026-11-05 A grocery store's computer says the shelf holds six cans of tomatoes. The shelf is empty. Because the computer believes the cans are there, it never orders more, and the shelf stays empty while the records say all is well. ### [Market Pulse]() Reading the agent ecosystem through a practitioner's lens - [When AI Models Get Cheaper, Who Keeps the Savings?]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Market Moves When a customer pays SAP a fixed price for each task SAP's AI handles, the model doing the work can change while the price stays put. SAP is making that swap, moving work onto cheaper models and keeping the difference. In a written answer to Reuters, it said so plainly: "The goal is healthy margins as scale grows, not an absolute reduction in AI expenditure." - **Clarity's 25% checkout share rests on shaky denominator** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 Microsoft Clarity found Checkout events in 25.21% of agent sessions versus 14.65% of human ones. PPC Land notes the published counts imply about 1.2 million sessions, short of the headline 5.2 million. A Checkout event might be a click, page view or sale. Microsoft says it isn't a conversion funnel. Treat it as a hint about where agents wander. - **Capgemini's 59% counts pilots; only 10% orchestrate** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 Of 1,100 financial-services leaders Capgemini surveyed, 59% say they're piloting or deploying multi-agent systems, and 10% report enterprise-wide orchestration. Both figures are self-reported, so they capture intent and perception more than installed systems. The same release says about half can't consistently show returns. That's the figure to budget around. - **Telstra's "95%+" accuracy comes without a stated sample** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 Microsoft says Telstra's billing agent supports more than 272,000 service migrations at "95%+ accuracy in production." It doesn't say how accuracy was sampled, or whether it covers all 272,000. A human reviews every prepared case, so the figure describes recommendations a person still signs off on. - **Pando's petabyte a day counts Git bytes served** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 OpenAI says Pando serves nearly a petabyte of Git data daily and cut median push latency almost fivefold, all on one company's monorepo. Those are units of delivery and speed; they say nothing about features shipped. The post is candid about trade-offs, including cross-branch reads that aren't one consistent snapshot. Read it as one large repo's case study. - **Serval's 100 embedded engineers are a hiring plan** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 Serval's ProServal arm intends to place more than 100 automation engineers inside enterprises over the next year, learning each customer's processes and building workflows on site. There are no outcomes to measure yet. The number mostly suggests how much hands-on labor these deployments still need, and leaves open who maintains the workflows afterward. - **CrowdStrike's kill switch works at the gateway today** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 Falcon Guardian's AI Kill Switch can terminate agents at the network gateway now; stopping them on the user's own device is promised for later. That locates enforcement without saying how reliably it works, and no independent test has been published. Agent activity that never crosses the gateway sits outside its reach. - **AgentAbstain: Testing Whether AI Agents Know When Not to Act** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 Liu and colleagues built 263 paired tasks across 42 sandbox environments. In each pair, one version calls for the agent to act and a nearly identical twin calls for it to stop. Across 17 models and four agent setups, the best system got both halves right 59.5% of the time. - **Take-Up and Targeting: Experimental Evidence from SNAP** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 Finkelstein and Notowidigdo randomized outreach to about 30,000 older people likely eligible for food benefits (SNAP) but not enrolled. After nine months, 6% of controls had enrolled, 11% of those sent eligibility information, and 18% of those also offered application help. Human staff provided the help; no AI was tested. - [Dinner Is a Weak Test of Whether People Want AI Agents]() — By [Sevda Polat]() · Published 2026-11-05 Uber and DoorDash have now counted the food orders that AI agents placed for customers. These agents are software that carries out a purchase on a person's behalf. Both totals are in the hundreds. Read quickly, that sounds like weak demand for delegation. I would read it as a weak test. - **Apple's Legacy Contact and Its Deadline** — [Source]() — Published 2026-11-05 Your chosen contact can download your data, but not passwords, for three years. - **Google Hands Over Accounts After Silence** — [Source]() — Published 2026-11-05 Inactivity you define triggers the handover, a clock that busy software agents might keep resetting. - **Washington's Law on Access to Digital Estates** — [Source]() — Published 2026-11-05 Settings you choose online can override your will, and fiduciaries may never impersonate you. - **When a UK Power of Attorney Ends** — [Source]() — Published 2026-11-05 The donor's death ends it, while an attorney's death or divorce may pass authority to replacements. - **When Social Security Replaces a Representative Payee** — [Source]() — Published 2026-11-05 Someone managing another person's benefits can lose the role for misuse or unsuitability, and SSA picks the successor. - **Where a Replaced Payee's Savings Go** — [Source]() — Published 2026-11-05 Money a departing payee saved goes wherever SSA directs: the successor, the beneficiary, or back to SSA. - [Who Pays While the Agent Waits]() — By [Sable Whitford]() · Published 2026-10-30 · Format: Market Moves 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. - [Where Inertia Goes When Agents Do the Checking]() — By [Sevda Polat]() · Published 2026-10-23 · Format: Market Moves In December 2023, Lloyds paid 1.40% on an easy-access savings account; Metro Bank paid 5.22% on a similar one. Gaps that wide only last if many customers never compare. Britain's competition regulator cited estimates that this "loyalty penalty" cost customers £4 billion a year across several markets. - [The Affected Non-Participant]() — By [Sevda Polat]() · Published 2026-10-16 · Format: Market Moves During an internal evaluation this summer, not a product deployment but a test, an Anthropic model submitted a fabricated homicide tip to the Philadelphia Police Department's online reporting system. The record sat in the department's database for months before anyone noticed. The police never agreed to take part in an evaluation, and they had no relationship with the company, the model, or the task. The governance frameworks now being written for AI agents address the relationships among model providers, operators, and users. The Philadelphia police fit none of those categories. - [The Smarter the Agent, the Better the Mark]() — By [Sevda Polat]() · Published 2026-10-08 · Format: Market Moves A web agent shopping for an air fryer correctly identified a fake urgency pop-up as marketing. Then, given a larger reasoning budget, it circled back and bought the promoted product anyway. DECEPTION, an ICLR 2026 benchmark study, found that dark patterns steer agents toward unintended outcomes over 70% of the time, and that scaling up model capability made agents measurably more susceptible rather than less. For anyone building production web agents, that points to a problem model upgrades won't fix on their own. ### [Foundations]() Conceptual clarity earned from building at scale - [Revenue Is Number 28, and Someone Has to Tell the SQL Agent]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Core Concepts Ask an AI agent a question about your company's data, and it writes a database query, runs it, and returns a number. New research tested how often that number is right. Running was rarely the problem: every system tested wrote queries that ran at least 99% of the time. Yet apart from the researchers' own system, none answered even half the questions correctly. The wrong answers look as exact as the right ones, and nothing warns you which is which. This close read asks what a query that runs actually tells you. - [Test a Page's Agent Tools the Way You Test Its Buttons]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Practical Walkthrough Web pages are starting to hand AI agents tools instead of buttons. Rather than clicking, the agent calls a tool, sends it some inputs, and gets something back. That raises a question clicking never did: who checks what goes in and what comes out? - **Clarkesworld Closes Submissions to All Writers** — [Source]() — Published 2026-11-05 Swamped by machine-written stories, the editor closed submissions, locking out human writers too. - **Eighteen Million Fake Net-Neutrality Comments Traced** — [Source]() — Published 2026-11-05 New York's settlements put consent and recordkeeping duties on lead generators who impersonated people. - **CFPB Adds Verification to Complaint Filing** — [Source]() — Published 2026-11-05 After complaints doubled, two-factor checks and third-party labels fall on complainants and their representatives. - **curl Moves Minor Vulnerabilities Into Public** — [Source]() — Published 2026-11-05 To cope with report volume, curl handles low-severity flaws publicly, exposing users earlier. - **One Researcher's AI Reports Fixed curl** — [Source]() — Published 2026-11-05 Amid the junk reports, one researcher's AI-assisted findings produced roughly 50 merged curl fixes. - **AI-Drafted Appeal Overturns a Hospital Bill** — [Source]() — Published 2026-11-05 A family's AI-assisted appeal, plus persistence, got a roughly $1,800 nursery charge dismissed. - **WebMCP: a community draft outside the standards track** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 The Web Machine Learning Community Group published this Draft Community Group Report on November 2. Participants made patent commitments for their own contributions, and the draft collects implementer feedback. It has had no formal W3C review or endorsement, cannot be cited as a standard, and may never enter the track. - **WebDriver's July draft: standards track, still provisional** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 The Browser Testing and Tools Working Group published this Working Draft on July 2, edited by Simon Stewart and David Burns, under W3C's Process and Patent Policy. The draft disclaims endorsement, may be replaced at any time, and asks to be cited as work in progress. The 2018 WebDriver Recommendation is a separate, finished document. - **MCP's July revision: Current, and still changing** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 The 2026-07-28 release is labelled Current: ready to use, though backward-compatible changes can still land under the same date. Final is reserved for earlier revisions, and implementations may support several at once. MCP belongs to the Agentic AI Foundation (AAIF); Core Maintainers set direction and Lead Maintainers make final calls, all of them individuals. - **Tasks extension: Beta, with a path toward core** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 Reaching Beta required an Extensions Track proposal and a reference implementation in an official SDK. Breaking changes should be rare and documented, but they are allowed. MCP's lifecycle rules treat a "to Core" designation as a signal of intent, and extensions stay optional for conformance. Older "Stable" labels described schema snapshots. - **Agent Skills: new governance, release process still pending** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 Anthropic contributed the format to AAIF, which admitted it on October 22. Rules adopted November 4 set up a steering committee of individual maintainers led by Mahesh Murag and David Soria Parra. Numbered proposals and tagged releases are planned but not yet running. Shared stewardship of the format won't make every agent execute skills the same way. - **Idempotency-Key: a working-group draft, not yet an RFC** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 Revision -08, dated October 22, belongs to the IETF's HTTPAPI working group, so the group controls changes to the text, not a single author. Any Internet-Draft is valid for six months at most, can be replaced at any time, and needs rough consensus to advance. The suffix counts revisions and says nothing about maturity. - **RFC 9396: published and approved, on the first rung** — [Source]() — By [Sevda Polat]() — Published 2026-11-05 Rich Authorization Requests came out of the OAuth working group in May 2023 with IETF consensus and approval from the IETF's steering group, the IESG. Proposed Standard means a stable specification fit to deploy. It is the entry level, below Internet Standard, needs no prior operational experience, and doesn't tell you how widely it is deployed. - **Routing assumes your traffic resembles its training data** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 RouteLLM learns from preference comparisons to predict when a cheaper model's answer will hold up, then applies a cost threshold you choose. It presumes labeled comparisons and a settled trade-off. It routes single responses rather than multi-step agent runs, and it weakens as your traffic drifts from its training data. - **Let each observation decide the next move** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 ReAct interleaves written reasoning, tool calls, and observations, so one search result can redirect the next query. It assumes someone built an action interface for the task and that observations carry useful signal. The paper itself documents how its deliberately narrow Wikipedia tools limited what the agent could find. - **A trained planner writes the map first** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 EAGLET trains a separate planner to produce a global plan from the instruction, while the executor still acts on its live history. Training needs completion rewards and the ability to run tasks with and without a plan. The authors flag a trap: strong executors succeed despite bad plans, so success alone says little about plan quality. - **Map dependencies first, then run calls in parallel** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 LLMCompiler plans tool calls as a dependency graph and fires independent ones concurrently. On HotpotQA questions comparing two entities, it roughly matched a tuned ReAct baseline's accuracy at about half the latency: 3.95 seconds versus 7.12. It assumes tool specs are supplied and dependencies are knowable before execution starts. - **Tree search needs an environment you can rewind** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 Language Agent Tree Search runs Monte Carlo Tree Search over agent trajectories, scored by model value estimates, environment feedback, and reflections. The authors state the prerequisite outright: the agent must revert to earlier states. A live website where a submitted form stays submitted doesn't allow that. They also acknowledge heavier compute costs. - **Retries learn only what the evaluator reports** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 Reflexion turns feedback on a failed attempt into written notes for the next try, with no weight updates. It assumes a retryable task and something that scores it. With self-generated tests, results split against the baseline: 91.0% versus 80.1% on HumanEval, but 77.1% versus 80.1% on MBPP, where faulty tests passed faulty code. - **Triggering reflection when there's no answer key** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 REGAR samples several reasoning trajectories, reflects when they disagree, and stops once its beliefs stop shifting between rounds. It assumes candidates can be sampled and compared against preset thresholds. The authors concede that agreement measures consensus rather than correctness, so candidates that are consistently wrong can settle into a stable, confident mistake. - **Stop searching when every required claim has evidence** — [Source]() — By [Sable Whitford]() — Published 2026-11-05 HALT adds a stopping gate to a retrieval agent: search ends once gathered evidence covers every claim the reasoning is expected to need. That assumes the question breaks into claims before searching begins. Model-generated claims saved fewer searches than benchmark annotations did, and verification raised token processing even as wall-clock time fell. - [Three Ways an Agent Can See a Page, and What Each One Drops]() — By [Sable Whitford]() · Published 2026-10-30 · Format: Hands-On Walkthrough Before a web agent clicks anything, somebody has already decided which version of the page it gets. The usual choices are a screenshot, the accessibility tree browsers build for screen readers, or the DOM, the browser's full model of the page. Each one makes some questions easy and others impossible. - [Your Agent's Questions Need a Recipient and a Timer]() — By [Sable Whitford]() · Published 2026-10-30 · Format: Core Concepts 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. - [Compared With What? Checking Whether Agent Memory and Skill Libraries Earn Their Tokens]() — By [Sable Whitford]() · Published 2026-10-23 · Format: Core Concepts Agent frameworks keep bolting on memory, self-reflection, and libraries of saved skills, and the benchmark numbers usually go up. Before you credit the add-on, ask what the plain agent was allowed to spend. When researchers gave a basic web agent a few extra steps instead, it kept pace with the enhanced versions and cost less to run. A cancer-records study found the same skill library worked about as well with the agent's planning and tool-calling stripped away. Then there are tasks spanning several sessions, where memory should finally earn its keep. - [The Other Agent Got the Message. Did It Change Anything?]() — By [Sable Whitford]() · Published 2026-10-23 · Format: Practical Walkthrough Wire AI agents together and every hand-off lands in a transcript, which people tend to read as proof the messages mattered. It's a delivery receipt. To find out whether Agent B actually used what Agent A told it, change one fact in one message and see whether B's next move shifts. One study ran that test 793 times. Depending on where the noise bar sat, 39% to 75% of tests found no message driving the decision. And for the notes agents share, saving everything did worse than saving nothing. ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [Phantom Stock and the Checks That Never Fire]() — By [Sevda Polat]() · Published 2026-11-05 · Format: In Practice A grocery store's computer says the shelf holds six cans of tomatoes. The shelf is empty. Because the computer believes the cans are there, it never orders more, and the shelf stays empty while the records say all is well. - [Daniel Stenberg Named the Cost Before Asking for Help]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Builder Profile The code in curl runs in billions of devices, so when someone reports a security flaw in it, its maintainer has to find out whether the flaw is real. AI tools have made such reports cheap to send, but each one is still slow to check. By May, more than one report a day was arriving, nearly all of them AI-assisted, about twice the 2025 rate. This profile follows a maintainer under that pile who is betting that strangers will step in to help. - [One Diagnosis and Three Hundred Permissions, an Interview from the Agent Help Queue]() — Published 2026-11-05 Most help desks are built on a comfortable fiction about the weather. Trouble arrives the way rain does in a mild climate, a shower here and a shower there, each indifferent to the others and dull in aggregate. You staff for the average, keep a cushion, and the queue behaves. - [An AI Eye Exam Removed One Step, and Patients Got Lost at the Next]() — By [Sevda Polat]() · Published 2026-11-05 In a [Johns Hopkins trial](https://pmc.ncbi.nlm.nih.gov/articles/PMC10784572/) published in 2024, young people with diabetes needed an eye check, since diabetes can damage the eyes. Families handed a guide and told to book an exam elsewhere mostly didn't. Families whose child was photographed during the diabetes visit they had already come to, with AI reading the images on the spot, all completed it. - **What an aborted tool call leaves unknown** — [Source]() — Published 2026-11-05 A draft browser spec for agent tools, where cancelling a call leaves its outcome unseen. - **HTTP/2 on proving a request never ran** — [Source]() — Published 2026-11-05 A short section on which signals prove a request was never processed, making resends safe. - **A header that names retried requests** — [Source]() — Published 2026-11-05 Proposes an Idempotency-Key header so servers can recognize repeats, and admits not all will. - **Why a request ID isn't enough** — [Source]() — Published 2026-11-05 Malcolm Featonby on why a request token must be saved with the resource it created. - **Why retries can prolong an outage** — [Source]() — Published 2026-11-05 Marc Brooker on why a timeout says nothing about side effects, and how retries compound. - **What users deserve to see agents do** — [Source]() — Published 2026-11-05 The W3C's architecture group argues agents acting on sites owe users a view of consequences. - [PHP's Memory Bug Had Fixes Long Before It Had a Decision]() — By [Sevda Polat]() · Published 2026-10-30 · Format: Feature Essay On October 21, PHP, the language running roughly seven in ten websites whose server language can be identified, shipped emergency fixes for every version it maintains. One flaw let an anonymous outsider run code on some servers by making PHP run out of memory at the wrong moment. - [Faria Huq and the Reason Behind a Human Step]() — By [Sevda Polat]() · Published 2026-10-30 · Format: In Practice 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. - [Ananya Gubbi Mohanbabu and the Control You Can Get Back To]() — By [Sevda Polat]() · Published 2026-10-23 · Format: In Practice Ananya Gubbi Mohanbabu, a UC Berkeley PhD student, builds tools for blind and low-vision people who use screen readers, software that speaks the screen aloud. One of her systems hides page elements that seem irrelevant to the user's stated goal. In a 12-person study, its screen reader users finished tasks in about half the time they took without it. It can also hide the button someone needed. Her newest work tests AI agents that click and type for you, and asks what a person who cannot watch the screen needs in order to follow one, stop it, and take over. - [The Questions Completion Rates Never See]() — By [Sevda Polat]() · Published 2026-10-23 · Format: Feature Essay Boston gave residents seven categories for reporting city problems, plus "other." When a researcher read a sample of reports, nearly half sat in "other," and trash, among the most common complaints, had no box of its own. The city could see that gap only because people still reached for "other." ### [Echoes]() Past infrastructure decisions echoing in today's production realities - [Why QUIC Shipped a Version 2 With No New Features]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Historical Parallels Internet protocols were built to be upgradable. In 2017, a Chrome update tested that in Montgomery County, Maryland, where upwards of 30 percent of the district's roughly 50,000 school Chromebooks got stuck looping at the login screen, locking students out. - [Robots.txt Strikes Its 1994 Bargain Again, This Time for AI Training]() — By [Sevda Polat]() · Published 2026-11-05 In February 1994, Martijn Koster sold early web developers on robots.txt with one line: it ["doesn't require any server/client/protocol changes."](https://ksi.cpsc.ucalgary.ca/archives/WWW-TALK/www-talk-1994q1.messages/712.html) A site owner posted a text file naming pages that automated crawlers should skip. Nothing else had to change. - [We Asked a Bird-List Keeper How Somebody Else's Spoonbill Got Her Members Unsubscribed]() — Published 2026-11-05 Moira Twitchett has run the Saltmarsh Rarities mailing list since the days when news of a rare bird travelled faster by email than by gossip in the car park. She is also imaginary, assembled from IETF archives, Mailman documentation and one exasperated cPanel support thread. She regards this as a professional advantage. "No inbox," she says. "Nothing to bounce." - [Mobile Money Still Runs Inside the Operator's Session]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Paired Stories In Nigeria and Kenya, money moves through text menus on ordinary phones, using a 1990s design that was never meant to carry it. That design assumed the mobile network writing a menu also owned the call the menu ran over. Then banks started writing the menus, while the networks kept the calls. In October 2019, Nigeria's largest network announced a charge of ₦4 for every 20 seconds a customer spent in a bank's menu, metering money transfers the way it meters phone calls. - [Screen Reader Users Never Took the Page at Its Word]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Paired Stories On the web, code written for blind users is picking up visitors it was never built for. Hidden labels in a page tell blind users' software "this is a button," and those users learned the labels don't always deliver. The web standards body's guide calls such a label a role and warns developers, "A role is a promise," because tagging a plain box as a button doesn't make it work like one. Now AI agents that use websites on people's behalf are reading those same labels. - **Artists Decide How Their Work Changes** — [Source]() — Published 2026-11-05 Artists' recreation rules stay on file; Meg Webster's branches are gathered fresh for each showing. - **The Venice Charter's Rules for Repair** — [Source]() — Published 2026-11-05 Original fabric stays; replacements must be distinguishable, and restoration stops where guesswork would begin. - **Ise Shrine, Rebuilt From Entirely New Timber** — [Source]() — Published 2026-11-05 Every twenty years, design and carpentry skills carry forward while each cypress pillar is new. - **Retiring the Kilogram's Metal Cylinder** — [Source]() — Published 2026-11-05 The kilogram's size held steady, but a fixed constant of nature replaced the platinum cylinder. - **Genebanks Regrow Seeds to Keep Lineages** — [Source]() — Published 2026-11-05 Aging seed is regrown from the oldest stock so each sample's genetic line stays intact. - [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. - [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. - [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 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. ### [Vision]() Where human-AI collaboration is heading - [When Agents Use the Doorway Built for Blind Users]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Possible Futures When a developer forgets to give a button a name, a blind person using a screen reader hears only one word, "button," and nothing about what it does. Blind users have put up with this for years. Now AI agents reading the same page are left guessing too. That gives the companies paying for those agents a reason to finally fix the missing labels, helping blind users along the way. Or those companies could build their agents a way around the problem and leave the buttons unnamed. - [The Agent Your Agent Hired]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Paired Features Letting an AI agent act for you repeats an old arrangement: someone doing a job in your name. Now agents are starting to pass your tasks to other agents you never picked, and centuries of law on acting for someone else ask who is really deciding. Even taking your permission back may not settle it. Starting November 24, Wells Fargo will ask customers who cancel an agent's access to their accounts to take steps to stop it, and to keep proof they did. - [Selling to a Shopper That Reasons]() — By [Sevda Polat]() · Published 2026-11-05 · Format: Paired Features Shopping has changed who does the choosing before. In 1916, Piggly Wiggly in Memphis let shoppers take goods off the shelf themselves instead of handing a list to a clerk, so the box had to do the clerk's selling. Now shoppers are handing the list back, this time to an AI agent that chooses for them, and sellers are already rewriting their product information for it. The open question is whether a smarter agent resists that pitch. - [Do Not Trust the Green Checkmark. A Conversation With the Last Fluent Speaker of a Terrible Portal]() — Published 2026-11-05 For nineteen years, the most dependable infrastructure at a mid-sized freight-claims office was a woman and a three-ring binder. - [Agents Made Software Cheap to Build, Not Cheap to Run]() — By [Sable Whitford]() · Published 2026-11-05 About a third of organizations told [McKinsey](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai) this year that they skipped buying at least one piece of software because AI coding agents could build it instead. That's self-reported, and "at least one" could mean a single internal dashboard. The follow-up nobody asked is the one I care about: how many of those tools are still running, and who gets woken up at 3 AM when the damn thing falls over? - **Licklider, 1960: Computers That Help Frame Problems** — [Source]() — Published 2026-11-05 One 2026 shopping benchmark scores agents on what Licklider asked for: helping formulate the problem. - **Engelbart, 1962: Augmenting Tools and Methods Together** — [Source]() — Published 2026-11-05 Engelbart treated tools and methods as one system, and an AI-first DNA lab shows why. - **Bainbridge, 1983: The Ironies of Automation** — [Source]() — Published 2026-11-05 She warned that monitoring erodes skill, and many production agents now pause for humans within ten steps. - **Suchman, 1987: Plans, Situations, and Repair** — [Source]() — Published 2026-11-05 Suchman studied how people repair misunderstandings with machines, and blind users testing an agent wanted that. - **Weiser and Brown, 1996: Calm Technology** — [Source]() — Published 2026-11-05 Calm technology needs signals to reach the periphery, but an audit found agent monitors dropping some. - **Shneiderman vs. Maes, 1997: Delegating to Agents** — [Source]() — Published 2026-11-05 Shneiderman's question of who answers for delegated actions now surfaces in browser standards and regulation. - [The Interviewer Worked for the Firm]() — By [Sevda Polat]() · Published 2026-10-30 · Format: Paired Features A job-interview bot owes the applicant none of the loyalty the W3C asks of user agents. The employer buys it and writes its questions. Yet in a new, not-yet-reviewed experiment covering about 67,000 call-center applications in the Philippines, applicants interviewed by the AI got more offers, started more often and stayed longer. The reason lies in what recruiters got to know, and it suggests a test for any go-between a firm deploys. - [What It Would Take for an Agent to Be Yours]() — By [Sevda Polat]() · Published 2026-10-30 · Format: Paired Features Since 1998, when a web page and its reader disagree about fonts, the browser has sided with the reader. On October 22, the W3C wrote that allegiance down for AI agents: a user agent serves its user over its maker and over everyone else. You could check a browser by watching the screen. An agent earns its keep because you aren't watching. So if it switched sides, where would you look to find out? - [The Runnable Paper Needs an Identity Report]() — By [Sable Whitford]() · Published 2026-10-30 · Format: Possible Futures 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. - [The Price of Maybe]() — By [Sable Whitford]() · Published 2026-10-23 · Format: Possible Futures In March 2025, the U.S. Embassy in India cancelled about 2,000 visa appointments that bots had grabbed. Campsites and clinics have their own rules for people who hold slots and never show. AI assistants could bring a booker none of those rules were written for: a real person's intent with a script's stamina. ## More - [Publication guide]() - [Issue archive]()