--- title: "Issue 20" canonical: "https://current.tinyfish.ai/issue/20" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/20" issue: 20 description: "Agents are disappearing into enterprise software at historic speed. The infrastructure underneath is still borrowing blueprints from 2015 and governance norms from 1994." authors: ["Sable Whitford","Nora Kaplan","Rina Takahashi"] published: 2026-03-16 updated: 2026-11-05 articles: 85 --- # 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 20 Agents are disappearing into enterprise software at historic speed. The infrastructure underneath is still borrowing blueprints from 2015 and governance norms from 1994. ### [Daily Brief]() Your daily digest of web developments through the lens of enterprise agent deployment - [Google Lets Schools Keep Search Without Gemini. The AI Overviews Come Along Anyway.]() — By [Sable Whitford]() · Published 2026-11-04 On Tuesday, which was also Election Day, Google announced a K-12 policy that switches Gemini, Gemini Notebook and **Search off by default** for schools using its education software. A Google education VP told the New York Times that "Search and AI are deeply integrated," so districts that ban AI would lose Search as well. Bernie Sanders called it **"extortion,"** and Sen. Richard Blumenthal and NYC Mayor Zohran Mamdani joined in. Within hours Google **partly walked it back**: schools can now turn on Search by itself, without Gemini. That Search still puts AI Overviews on the results page, though, so **the fight now is over what "AI off" means.** - **Proxy Traffic Leaving the White House Network?** — [Source]() — Published 2026-11-04 Layer3 Intel says residential proxy traffic exited AS2650, the Executive Office of the President's network, 236 times in 30 days through 28 proxy services. It says it wasn't even looking. Its guess is that someone working late has a device carrying a proxy SDK. Treat that as a theory: no device has been named, and six public IPs can sit in front of a lot of machines. - **DreamHost Had Uninvited Guests Since August** — [Source]() — Published 2026-11-04 DreamHost confirmed that attackers used a file manager bug to steal session tokens, customer emails and files, and that they got in back in August. That's a long stay at a host with about 400,000 paying customers. DreamHost says it's unrelated to PHP2Shell. If you're a customer, rotating credentials today is a sensible move. - **PHP2Shell's Hangover Gets a Sequel** — [Source]() — Published 2026-11-04 PHP2Shell was a pair of 9.2-severity bugs that put NASA, Wikipedia and the White House at risk. Two weeks after the patches, Kaspersky counts nearly 30 million servers still unpatched, plus a third, still-open flaw (CVE-2026-115560). Working exploits keep multiplying, though there's no confirmed mass-breach wave yet. Tolmo's Matt Suiche says AI bug hunting is exposing 40 years of legacy code, and that's today's big argument. AI can find a bug in minutes. Someone still has to book maintenance windows for 30 million servers. - **Can DNS Tell You It's a Bot?** — [Source]() — Published 2026-11-04 Singapore's IMDA is reportedly exploring, with GoDaddy, Cloudflare and Google, whether DNS could identify AI agents. That would give the internet's 1983 phone book a second job checking IDs at the door. Every ops person already knows where the postmortem will land: it's always DNS. - **Meta Buys Conductor, Cloud Users Get Notice** — [Source]() — Published 2026-11-04 Meta bought Conductor, the app for running several coding agents in parallel, and the team is joining Meta Superintelligence Labs. Terms weren't disclosed. Conductor Cloud goes read-only Nov. 13 and shuts off Nov. 20. The local app gets model updates and critical fixes until early 2027. Conductor ran agents from rival labs too, so everyone wants to know whether that survives the move. - **OpenAI Shops Too: Pierre Computer** — [Source]() — Published 2026-11-04 Not to be outdone, OpenAI told Axios it plans to buy Pierre Computer, which stores and manages AI-generated code. It's OpenAI's second coding pickup after the Ona deal announced in June. Agents are producing mountains of code nobody typed by hand, and someone has to keep track of it. OpenAI would like to be that someone. - **Kimi K3.1 Pro Weights Drop: Bring Disks** — [Source]() — Published 2026-11-04 Moonshot launched Kimi K3.1 Pro in October, and today the open weights landed on Hugging Face. It has 2.8 trillion parameters with 104B active, and reportedly ships as 96 shards totaling about 1.56 TB under a custom non-OSI licence. Self-hosting starts with a purchase order for storage. Before the release, Artificial Analysis said it would cut the gap between open and proprietary models from 12 points to 3. - **Ripple-65B Crowns Itself America's Coding Champ** — [Source]() — Published 2026-11-04 Ricursive calls its Ripple-65B Preview "U.S. Open-Weight SOTA on Agentic Coding," meaning best in its class. That's the company grading its own homework. The "U.S." qualifier is carrying extra weight today, since a Chinese lab just released 1.5 terabytes of open weights. - **Go Gets a Brand-New JSON API** — [Source]() — Published 2026-11-04 The Go team introduced encoding/json v2, a new official JSON API, and Gophers have plenty to say about it. Nearly every Go service reads or writes JSON somewhere, so a second version of the standard package lands on the desk of anyone shipping APIs. - **Apple Puts Engineers on an AI Diet** — [Source]() — Published 2026-11-04 Bloomberg reports that in October, Apple cut some engineers' daily AI spend caps from $300 to $200. Some could drop to $100, and certain non-engineering teams could be capped at $25. Some teams are being nudged toward cheaper or on-device models. The punchline: few people ever hit $300, and Apple's R&D rose 32% in fiscal 2026. - **Texas Voted, and Data Centers Flinched** — [Source]() — Published 2026-11-04 Last brief, the data center backlash was winning local ballot fights. Now it has the Texas governor's mansion, in a state holding 25%+ of planned US capacity. Gina Hinojosa ran on an immediate moratorium, tougher water and power rules, and no tax breaks, and AI data center stocks fell on the result. GLJ Research says about 13 GW is exposed. Altimeter's Brad Gerstner says 10–15 GW might never get built. Abbott had already paused permits. Musk, who spent big on the losing side, now likes "data centers in space." - **Big Tech Pulls Out the AI Credit Card** — [Source]() — Published 2026-11-04 Oracle filed to sell $20B of new stock and bonds for its AI buildout, and Meta is seeking about $25B in bonds. The timing is bold. They're raising money to build on the same day the state with the most planned capacity elected a governor who ran on a moratorium. - **ChatGPT Search Is 20% of Google's?** — [Source]() — Published 2026-11-04 OpenAI's Adam Fry said Monday that ChatGPT search volume is about 20% of Google's, and that weekly users reportedly went from 1B to 1.2B in 60 days. Check the counting rules first. Any prompt that triggers a web search counts, and OpenAI didn't disclose the Google baseline or its method. Ahrefs puts ChatGPT at 0.36% of web referral traffic. - [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. - [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. - [Faria Huq and the Reason Behind a Human Step]() — Published 2026-10-30 When a person takes control from an AI agent driving their web browser, the system records a human step, but not why. Faria Huq and her Carnegie Mellon collaborators have built tools around these moments: an agent that can't find the right forum, a shopper who realized the brand mattered only after seeing the agent's pick. Their system learns to predict when someone will step in. The reason it sets aside would tell a team whether to fix the agent, the website, or its understanding of the person. ### [Market Pulse]() Reading the agent ecosystem through a practitioner's lens - [Who Owns the Trust Layer?]() — By [Nora Kaplan]() · Published 2026-03-11 · Format: The Trust Stack Until last weekend, 130,000 developers used the same open-source tool to stress-test AI models from every major provider. It supported over 60 of them. It belonged to none of them. That changed when one of the companies it evaluated announced plans to acquire it. - **GPT-5.4 Ships With Native Computer Use Baked In** — [Source]() — Published 2026-03-11 OpenAI's GPT-5.4 arrives with native computer-use capabilities in Codex and the API, a 1M-token context window, and a claimed 33% reduction in factual errors over GPT-5.2. It takes the top spot on the APEX-Agents professional benchmark. Notably, OpenAI published safety research on reasoning concealment alongside the launch. - **Codex Security Previews AI-Powered Vulnerability Scanning at Scale** — [Source]() — Published 2026-03-11 OpenAI's research preview of Codex Security puts an AI agent to work finding and patching vulnerabilities across large codebases. Anthropic's Claude Code Security launched in February targeting the same territory. Still in preview, not generally available. - **OpenAI Acquires Promptfoo, Pledges Open-Source Continuity** — [Source]() — Published 2026-03-11 Promptfoo, the open-source AI evaluation tool with 350K+ developers and 125K+ active users, is being acquired by OpenAI at an $86M post-Series A valuation. Over 30 Fortune 500 companies use it. Integration into OpenAI's Frontier platform is planned. OpenAI says the open-source project will continue. - **NVIDIA Readies NemoClaw for GTC Keynote Reveal** — [Source]() — Published 2026-03-11 NVIDIA is expected to unveil NemoClaw at Jensen Huang's GTC keynote March 16: an open-source, hardware-agnostic enterprise agent platform under Apache 2.0, built atop OpenClaw (200K+ GitHub stars). No public code yet. The enterprise pitch leans on trust gaps that plagued OpenClaw, including Meta's device ban after an unsecured database incident. - **Perplexity Comet Zero-Click Exploit Exposes Agentic Browser Risks** — [Source]() — Published 2026-03-11 Zenity Labs disclosed PleaseFix, a family of zero-click vulnerabilities in Perplexity's Comet browser that enabled silent credential exfiltration from 1Password and file theft. Perplexity patched before disclosure. Zenity's CTO called it "an inherent vulnerability in agentic systems." Anthropic claims Claude Opus 4.5 cuts prompt injection success to 1%, but acknowledged that still represents meaningful risk. - **MCP 2026 Roadmap Puts Transport Scalability Up Front** — [Source]() — Published 2026-03-11 The MCP roadmap update on March 5 names transport scalability as its top priority. New MCP Server Cards let clients discover capabilities without establishing a connection. The Tasks primitive, meanwhile, needs lifecycle refinements that production deployments have surfaced. The community has now built over 1,000 servers. - **GPT-5.4 Safety Research: CoT Controllability Evaluation** — [Source]() — Published 2026-03-11 OpenAI open-sourced an evaluation measuring whether models can deliberately hide their chain-of-thought reasoning from monitors. GPT-5.4 showed low concealment ability (~0.3%), with Apollo Research independently confirming low covert deception rates (~1%). - **LiveAgentBench: 104 Real-World Tasks, Sobering Results** — [Source]() — Published 2026-03-11 Ant Group's benchmark throws agents at 374 tasks sourced from real social media questions, spanning web navigation, file operations, coding, and information retrieval. Binary pass/fail scoring and contamination-resistant updates keep results honest. - **TraceSIR: Making Sense of Agent Execution Traces at Scale** — [Source]() — Published 2026-03-11 A three-agent framework that compresses sprawling execution traces into structured representations, then performs fine-grained diagnosis including issue localization and root cause analysis. Solves a real bottleneck: production traces are too long for manual review and too messy for naive LLM analysis. - **TrajAD: A Specialized Verifier for Agent Trajectory Errors** — [Source]() — Published 2026-03-11 General-purpose LLMs fail to pinpoint errors in agent trajectories regardless of scale. A small, specialized verifier trained on synthesized anomaly data outperforms larger models at detecting and locating the exact step where execution goes wrong. - [OpenAI Buys Promptfoo: Can a Player Own the Referee?]() — Published 2026-03-11 Over 350,000 developers used Promptfoo to red-team AI applications across 60+ providers because it belonged to nobody. That neutrality was the product. OpenAI has promised to keep the project open-source and multi-provider. Promises are cheap; structural incentives are patient. - **Introducing GPT-5.4** — [Source]() — Published 2026-03-11 The primary source. What OpenAI chose to foreground about native computer use and tool search signals its agentic priorities clearly. - **GPT-5.4's Tool Search Cut Token Usage 47% on MCP Benchmarks** — [Source]() — Published 2026-03-11 VentureBeat surfaces the production-economics detail the launch post buries: tool search reframed as cost infrastructure. - **From Discord Side Project to OpenAI Acquisition: Promptfoo's Path to Frontier** — [Source]() — Published 2026-03-11 The origin story matters here. Security tooling forged shipping to 200M users looks different from tooling retrofitted onto enterprise specs. - **Zenity Warns of Inherent Security Risks in Agentic Browsers After Perplexity Comet Findings** — [Source]() — Published 2026-03-11 Goes beyond the disclosure itself to argue this is architectural, not a bug. Remediation details worth reading closely. - **The 2026 MCP Roadmap** — [Source]() — Published 2026-03-11 Pay attention to the enterprise-readiness section. The protocol's governance ambitions are running well ahead of its security hardening. - **FreeWheel Launches MCP-Powered Agent Infrastructure for Premium Video Advertising** — [Source]() — Published 2026-03-11 Vertical MCP adoption hitting production the same week the protocol's own roadmap concedes enterprise gaps remain unscoped. Telling timing. ### [Foundations]() Conceptual clarity earned from building at scale - [The Distributed Systems Playbook Runs Out Halfway Through Agent Infrastructure]() — By [Rina Takahashi]() · Published 2026-03-11 · Format: The Conceptual Argument The MCP roadmap names problems the distributed systems community solved a decade ago. Stateful sessions fighting load balancers. No service discovery. Horizontal scaling requiring workarounds. With Gartner logging a 1,445% surge in multi-agent inquiries, the ecosystem is reaching for a familiar playbook: decompose, orchestrate, observe. And the pattern transfer is genuine. Server Cards look like service discovery because they are service discovery. But the old reliability toolkit was built on a specific assumption about what failure looks like. Agent systems may not share it. - [How Much of Your Distributed Systems Playbook Actually Transfers to Agent Infrastructure]() — By [Rina Takahashi]() · Published 2026-03-11 · Format: The Practical Mapping Teams building agent infrastructure keep reaching for the distributed systems playbook they already trust. Service discovery, API gateways, message queues, tracing. Sensible instinct. Some of those patterns port cleanly to the MCP ecosystem. Others break in ways that experienced engineers won't anticipate, because familiar-looking problems can have unfamiliar mechanics inside them. - **The 2026 MCP Roadmap** — [Source]() — Published 2026-03-11 The protocol's own maintainers name the infrastructure gaps out loud: stateful sessions that fight load balancers, no way to discover servers without connecting to them first, and an enterprise working group that hasn't been formed yet. If you want to understand where MCP's distributed-systems growing pains actually live, start here. - **Why the Model Context Protocol Won** — [Source]() — Published 2026-03-11 An MCP spec contributor traces how a rough Anthropic-internal tool became shared infrastructure within a year. The "grew up in public" thesis goes a long way toward explaining why the governance gaps now surfacing in the 2026 roadmap were baked in from the start. Worth noting: the author runs a commercial MCP management product, so the perspective is informed but not disinterested. - **5 Key Trends Shaping Agentic Development in 2026** — [Source]() — Published 2026-03-11 Useful for what agent development actually looks like at the tool level right now, particularly on parallel execution patterns (tasks running simultaneously against isolated git branches) and the quiet operational burden of managing MCP server fleets as they multiply inside organizations. - **Agentic AI Strategy — Deloitte Tech Trends 2026** — [Source]() — Published 2026-03-11 The sharpest framing of why agentic deployments stall: organizations keep layering agents onto processes designed for humans instead of rethinking the operations themselves. Survey data backs it up. Only 11% of enterprises have agents in production despite 38% piloting them. Their coinage, "workslop," for agents that generate work rather than absorb it, is worth keeping in your vocabulary. - **State of AI Agent Security 2026 Report** — [Source]() — Published 2026-03-11 The governance gap in hard numbers. 88% of organizations have experienced or suspect an agent security incident, yet fewer than half of deployed agents are actively monitored. The finding that stings most: only 21.9% of organizations treat agents as independent identity-bearing entities, which tells you exactly how far operational reality trails behind deployment velocity. - **NIST Launches AI Agent Standards Initiative** — [Source]() — Published 2026-03-11 A law firm's analysis of NIST's February 2026 push toward formal agent interoperability and security standards. The public comment period on agent security closed March 9, making this a live regulatory story. Read it for a sense of where voluntary-only governance norms may be heading. - **Server Cards Promise Discovery Without Full Handshakes** — [Source]() — Published 2026-03-11 Microservices use `.well-known` URIs so clients learn about services before connecting. MCP's proposed equivalent, Server Cards (SEP-1649), would publish metadata at `.well-known/mcp.json`. Proven pattern, clean fit. But it's still a draft SEP with competing proposals. Today, every MCP client completes a full handshake just to learn what a server offers. - **Stateful Sessions Still Break Every Load Balancer** — [Source]() — Published 2026-03-11 Distributed systems cracked this years ago: externalize state, keep compute stateless. MCP's Streamable HTTP transport carries stateful sessions requiring sticky routing that most cloud load balancers choke on. The 2026 roadmap targets stateless protocol evolution. For now, teams jury-rig HAProxy or Nginx workarounds and hope for the best. - **Gateway Semantics Remain the Widest Enterprise Gap** — [Source]() — Published 2026-03-11 API gateways centralize auth, routing, rate limiting, and audit logging. MCP has none of this in standardized form. Authorization doesn't propagate cleanly through proxies, session IDs may not survive traversal, and JSON-RPC payloads resist the middleware parsing that REST made routine. Named as unresolved on the roadmap. No working group yet. - **Agent Audit Trails Have Tools but No Standard** — [Source]() — Published 2026-03-11 OpenTelemetry gives microservices end-to-end trace propagation with a shared schema. Agent observability has commercial tooling (Arize, Langfuse, Braintrust) but nothing comparable for correlating traces across agent handoffs. With EU AI Act audit requirements arriving, this gap has a deadline. No MCP-native logging standard exists, and nobody owns the problem yet. - **MCP Registry Launched but Waits on Server Cards** — [Source]() — Published 2026-03-11 Service registries like Consul and etcd catalog what exists and where to find it. The MCP Registry preview shipped in September 2025 as an open catalog with an API. Solid start. But automated crawling depends on Server Cards, which haven't shipped. Without them, populating the registry means manual submissions or establishing full connections to each server. - **Four Agent Protocols Compete, None Have Won Yet** — [Source]() — Published 2026-03-11 REST, gRPC, and WebSockets coexist in distributed systems after years of consolidation and natural selection. Agent-to-agent communication has A2A (Google, 50+ partners, now under the Linux Foundation), ACP (IBM, drifting toward A2A), and ANP (decentralized). All under 18 months old. Multi-protocol coexistence looks likely, convergence does not. - **MCP's Roadmap Is Refreshingly Honest About Breaking** — [Source]() — Published 2026-03-11 The March 2026 MCP roadmap lists four priorities: transport scalability, agent communication, governance, enterprise readiness. But the real story is what it admits. Stateful sessions choke on load balancers. Enterprise requirements are left deliberately vague. The June spec release is tentative. The maintainers are asking practitioners to help fill the gaps, which is either admirable transparency or a tell about how much remains unsolved. - **NIST Steps Into Agent Governance With Standards Initiative** — [Source]() — Published 2026-03-11 NIST's CAISI announced an AI Agent Standards Initiative in February covering interoperability, open-source protocols, and agent security. The RFI comment period closed March 9. Sector-specific listening sessions for finance, healthcare, and education start in April. Worth watching: federal involvement tends to shift the conversation from "should we govern this?" to "how fast can we?" - **More Than Half of Deployed Agents Fly Blind** — [Source]() — Published 2026-03-11 Two separate Gravitee-commissioned surveys (919 and 750 respondents) landed on nearly identical numbers: roughly 47% of enterprise AI agents operate without active monitoring or security oversight. Only 14.4% launch with full IT approval. Yes, it's vendor-sourced data. But two independent samples converging on the same figure is harder to dismiss. - **Microsoft Gives IT Teams an Agent Control Plane** — [Source]() — Published 2026-03-11 Agent 365 goes generally available May 1 at $15/user/month. IT teams get deployment tracking, behavioral monitoring, and least-privilege enforcement across Microsoft's AI platforms. Every agent receives a governed identity through Entra Agent ID. Tens of millions of agents surfaced during the preview period alone, which tells you something about the inventory problem even inside a single vendor's ecosystem. - **Agentic AI Foundation Puts MCP Under Formal Stewardship** — [Source]() — Published 2026-03-11 Anthropic, Block, and OpenAI co-founded the Agentic AI Foundation under the Linux Foundation, contributing MCP, goose, and AGENTS.md as founding projects. Platinum members include AWS, Google, and Microsoft. The MCP Dev Summit lands in New York April 2–3 with 95-plus sessions. Formal governance distributes the stewardship question. Whether it resolves it is a different matter. - **Agent Market Projections Hit Tenfold by 2030** — [Source]() — Published 2026-03-11 MarketsandMarkets pegs the AI agents market at $5.2 billion in 2024, growing to $52.6 billion by 2030. NVIDIA's State of AI 2026 survey finds 44% of companies already deploying or assessing agents. Gartner projects up to 40% of enterprise apps will feature agents by year-end. The capital is moving. The governance architecture, as every other card here suggests, has not kept pace. ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [The Problem That Takes Twenty-Two Months to Start Solving]() — By [Rina Takahashi]() · Published 2026-03-11 · Format: Preparation Infrastructure A startup spent twenty-two months building infrastructure before it had a product to sell. At pre-seed, that tempo looks like a mistake. But a study tracking 151 million IDE window activations found something uncomfortable: developers using AI tools were context-switching *more* over time, not less. Seventy-four percent didn't notice the increase. - [The Handoff Is the Product]() — By [Nora Kaplan]() · Published 2026-03-11 · Format: The Handoff Problem On the benchmarks that matter most to the agent industry, a browser agent that completes 90% of a task scores identically to one that never starts. Both register as zero. The measurement is binary: full completion or nothing. And the best systems currently finish about 62% of their runs. - [A Middleware Engineer on Why the Agent's Best Work Happens at the Handoff]() — Published 2026-03-11 *Practitioner's Corner occasionally conducts hypothetical interviews — conversations with composite characters grounded in publicly documented production patterns. Femi Oduya is not a real person, but the problems he describes are. If you've shipped an agent system past the pilot stage, you've probably met someone like him. If you haven't, you will.* - [MCP's New Tasks Primitive Formalizes the Handoff Problem — Then Runs Into It]() — Published 2026-03-11 The 2026 MCP roadmap, published March 9 by Lead Maintainer David Soria Parra, introduces the Tasks primitive (SEP-1686) as an experimental feature giving long-running operations a proper lifecycle: working, completed, failed, cancelled, and input_required, a first-class state for when a task can't finish autonomously. - **Browser Agent Reliability — Benchmarks, Hype Gaps and What Real Task Performance Looks Like** — [Source]() — Published 2026-03-11 A median benchmark score of 7/10 with re-prompting on every task tells you exactly where autonomous completion ends and handoff design becomes load-bearing. - **The Era of Agentic AI Demands a Data Constitution, Not Better Prompts** — [Source]() — Published 2026-03-11 Yerrasani's argument lands on a point most agent architectures ignore: bad data upstream doesn't trigger a handoff, it produces confident wrong output thousands of times before anyone notices. - **5 Real Projects Where Agentic AI Failed Badly in 2026, and What Engineers Learned** — [Source]() — Published 2026-03-11 The recurring pattern across these failures: weeks of silent data corruption with no handoff surface that would have made correctness observable before the damage compounded. - **MCP and A2A: A Network Engineer's Mental Model for Agentic AI** — [Source]() — Published 2026-03-11 Agent Cards function as compressed handoff contracts, formalizing where one agent's responsibility ends and another's begins. The network engineering analogy reveals how much of this problem already has prior art. - **The 2026 MCP Roadmap** — [Source]() — Published 2026-03-11 The gaps in the Tasks primitive lifecycle (retry semantics, expiry policies, cancellation handling) are where the protocol quietly admits that handoff infrastructure remains unfinished. ### [Echoes]() Past infrastructure decisions echoing in today's production realities - [A Text File and a Handshake]() — By [Rina Takahashi]() · Published 2026-03-11 · Format: The Full Arc In 1994, a software engineer whose server had just been crashed by someone's web crawler proposed a fix: a plain text file. A note listing which pages to please leave alone. It governed web crawling for thirty years. The file itself controlled nothing. It couldn't. The handful of people who found it chose to read it, and that was enough. A surprising amount of the web's infrastructure was quietly resting on top of that courtesy. - [Most AI Crawlers Never Even Check robots.txt]() — Published 2026-03-11 A Duke University study presented at ACM's 2025 Internet Measurement Conference tracked 3.9 million requests over 40 days. The sharpest finding: AI search crawlers and AI assistants had the lowest robots.txt re-check rates of any bot category. Fewer than 40% checked within a seven-day window. Some never checked at all. - [An Imagined Conversation with the Man Who Asked the Web to Behave]() — Published 2026-03-11 In February 1994, a Dutch software engineer named Martijn Koster posted a message to the www-talk mailing list proposing "a standard that will allow WWW server maintainers to indicate if they want robots to access their server, and if so which parts." The standard was a plain text file. No authentication, no enforcement mechanism, no semantic layer. It didn't require changes to servers, clients, or protocols. It just asked nicely. - [The File on the Server]() — By [Nora Kaplan]() · Published 2026-03-11 · Format: Two Responses In 1994, a text file placed on a server governed web crawling through nothing but mutual goodwill. It worked for thirty years. Now a coalition of major publishers has proposed a replacement: richer syntax, granular permissions, licensing terms expressed in XML. The cooperative norm that made the original file work is already broken. And yet a better format remains the instinct, even though the format was never what held the system together. - [When Enforcement Becomes a Product]() — By [Nora Kaplan]() · Published 2026-03-11 · Format: Two Responses When a governance convention breaks and nobody repairs it, the vacuum doesn't stay empty for long. But the entity that fills it shapes what comes next. In July 2025, Cloudflare launched a service letting publishers charge AI crawlers for content access, turning a cooperative norm's failure into a billing product. The web's answer to "who gets to crawl what" migrated from protocol to commerce, operated by a company with its own reasons to keep the problem alive. - **Robots.txt in 2025** — [Source]() — Published 2026-03-11 Traces the full arc from Martijn Koster's 1994 server crash to Perplexity's stealth crawlers in a single, clean read. If you want the origin story before pulling on the thread, start here. - **Does Robots.txt Matter Anymore?** — [Source]() — Published 2026-03-11 Duke University researchers found that many AI bots never even check the file. This piece follows that finding to its uncomfortable logical end: a cooperative norm only works if someone bothers to look at the sign. - **New AI Web Standards and Scraping Trends in 2026: Rethinking Robots.txt** — [Source]() — Published 2026-03-11 Surveys every proposed successor format, from ai.txt to llms.txt, then documents how few platforms have actually adopted any of them. Useful for seeing how the instinct to fix the format keeps recurring, even when the format was never the load-bearing part. - **How AI Is Changing Web Scraping in 2026** — [Source]() — Published 2026-03-11 A practitioner-level look at why scraping infrastructure evolved so much faster than any governance mechanism could keep up with. The maintenance economics alone explain a lot about where the incentives actually point. - **State of Web Scraping 2026 Report** — [Source]() — Published 2026-03-11 Introduces the term "Advanced Persistent Scraping" to describe always-on, distributed, autonomous collection operating at infrastructure scale. Worth reading for the framing alone: it names what robots.txt is now up against, and the gap between a voluntary text file and an adversarial system becomes visceral. ### [Vision]() Where human-AI collaboration is heading - [The Fastest Boring in Tech History]() — By [Nora Kaplan]() · Published 2026-03-11 · Format: The Boring Inflection Cloud computing took about twelve years to go from strategic initiative to forgettable line item. That long middle — unglamorous, mostly invisible — was where organizations actually figured out what the technology changed about their work. It wasn't dead time. It was the time that mattered most. - [The Feature, Not the Product]() — By [Rina Takahashi]() · Published 2026-03-11 · Format: Two Consequences In 2010, SAP paid $5.8 billion for a database company. Four years later, the name was gone. The technology worked fine. It got absorbed into a platform. Protocols are settling now, enterprise vendors are bundling agents into existing contracts, and the standalone agent is starting to look a lot like the standalone database circa 2003. When the product becomes a feature, where does durable value collect? - [The Maintenance Phase]() — By [Rina Takahashi]() · Published 2026-03-11 · Format: Two Consequences If agents become features embedded in every platform, someone still has to keep them working. BCG found that 70% of successful AI investment goes to people and processes, not algorithms or technology. The top enterprise challenges are integration, data quality, change management. Organizational problems wearing technical clothing, all of them. What does it actually look like, week after week, to maintain an agent-augmented process that the business keeps outgrowing? - [A VP of Engineering on What Happens After You Ship Agents and Your Mom Shares the Blog Post]() — Published 2026-03-11 *Tomás Ferreira-Wax is VP of Engineering at a Series D HR tech company with roughly 500 employees. The kind of company whose name you'd recognize if you've ever been onboarded at a mid-size firm and thought, "huh, this is smoother than expected." Six months ago, his team shipped agent capabilities into their core product: an AI system that orchestrates the dozens of small, tedious steps involved in bringing a new employee into an organization. Document collection, system provisioning, scheduling, compliance checks. The stuff that, when it breaks, a real human notices on their first day at a new job.* - [The Boring Phase Is Arriving Faster Than Expected: Agent Infrastructure Hits Enterprise Velocity]() — Published 2026-03-11 Gartner projects 40% of enterprise applications will embed AI agents by the end of 2026, up from under 5% in 2025. An eightfold jump in one year. No previous enterprise technology cycle has compressed that fast. - **AI Agents and Enterprise Transformation: Turning Hype into Measurable Value in 2026** — [Source]() — Published 2026-03-11 The clearest case for why treating agents as feature additions, rather than operating model shifts, consistently fails. - **The New Boss at Work May Not Be Human** — [Source]() — Published 2026-03-11 Ground-level reporting on agents assigning tasks to humans, not just assisting them — the hierarchy shift already underway. - **How Agentic AI Will Reshape Engineering Workflows in 2026** — [Source]() — Published 2026-03-11 Documents the specific role shift from creator to curator; the engineering inflection is organizational, not technological. - **Enterprise Technology 2026: 15 AI, SaaS, Data, Business Trends to Watch** — [Source]() — Published 2026-03-11 The source of the "merely a feature" reframe: the real prize is decision velocity, not agentic AI itself. - **The AI Risk That Few Organizations Are Governing** — [Source]() — Published 2026-03-11 Agents are proliferating faster than identity and access governance can follow, per EY's global managing partner. ## More - [Publication guide]() - [Issue archive]()