--- title: "Issue 23" canonical: "https://current.tinyfish.ai/issue/23" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/23" issue: 23 description: "Thousands of tools, 85% reliability, a court win, 35% fewer entry-level jobs. Under every number, a gap that changes the story." authors: ["Sable Whitford","Nora Kaplan","Rina Takahashi"] published: 2026-04-06 updated: 2026-11-05 articles: 78 --- # 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 23 Thousands of tools, 85% reliability, a court win, 35% fewer entry-level jobs. Under every number, a gap that changes the story. ### [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. - [The Runnable Paper Needs an Identity Report]() — Published 2026-10-30 Paper2Agent, published in Nature in September, turns research papers into AI agents you can run. Hand over your data and the method executes, with no rebuilding it from prose. The authors want papers to carry an "agent availability" section saying whether a runnable version exists. - [Lebanon's Two Clocks and the Question Calendars Don't Ask]() — Published 2026-10-30 On March 26, 2023, Lebanon woke up on two clocks. The caretaker government had postponed daylight saving three days earlier. Churches and TV channels moved forward anyway, government offices stayed put, and most phones jumped on their own. Middle East Airlines printed every departure an hour earlier, so its planes still left at the same moment. That week made each calendar show what it had been storing: an exact instant, or a time on the wall that politics can move. Software assistants now make that choice without asking anyone. - [Your Agent's Questions Need a Recipient and a Timer]() — Published 2026-10-30 When an AI agent stops mid-task to ask a human something, it often reads as a failure of autonomy. In one study, letting coding agents ask about bug reports stripped of key details won back most of the lost ground, lifting solve rates from about 55% to 69.4%. But the "human" answering was another model, holding the full report and standing by. Real people are in meetings, or asleep. The design work is in the ask: who gets it, what it says, and what happens while it waits. - [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 - [The Court Punted. The Market Didn't.]() — By [Nora Kaplan]() · Published 2026-04-01 · Format: What Got Protected Judge Rita Lin's preliminary injunction in the Anthropic-Pentagon case runs forty-three pages. It is careful, narrow, and grounded in a constitutional principle that would bore most law students. By the time she issued it in late March, ten defense-tech companies under a single venture firm had already replaced Claude. Lockheed Martin had started swapping models. A rival signed its own Pentagon deal hours after the original blacklisting. - **Judge Blocks Pentagon on First Amendment Grounds** — [Source]() — Published 2026-04-01 A federal judge in California indefinitely blocked the Pentagon's supply chain risk designation against Anthropic. The 43-page ruling called it "classic illegal First Amendment retaliation" for bringing "public scrutiny" via the press. Anthropic was found likely to succeed on the merits. A final verdict could be months away. - **A Label Built for Huawei, Aimed Domestically** — [Source]() — Published 2026-04-01 The supply chain risk designation had only ever been used against foreign adversaries. Anthropic became the first American company to receive it. The label requires defense contractors, including Amazon, Microsoft, and Palantir, to certify they do not use Claude in military work. - **Defense Startups Drop Claude Within Days** — [Source]() — Published 2026-04-01 At least 10 portfolio companies at defense-focused J2 Ventures stopped using Claude for military work and started replacing it. Managing partner Alexander Harstrick: "Most of our companies are actively involved in large defense contracts and so are very strict in their interpretation of the requirements." - **Anthropic Drew Two Lines the Pentagon Wouldn't** — [Source]() — Published 2026-04-01 Anthropic wanted contractual assurances on two points: no fully autonomous weapons, no mass domestic surveillance. The Pentagon wanted unfettered access across all lawful purposes. Anthropic published "Where We Stand" laying out its position. Neither side moved. - **OpenAI Announced Pentagon Deal Hours After Blacklisting** — [Source]() — Published 2026-04-01 Hours after the Pentagon designated Anthropic a supply chain risk, OpenAI announced it had agreed to DoD terms. Sam Altman later called the timing "sloppy," saying the company "shouldn't have rushed." OpenAI subsequently renegotiated to include surveillance limitations. - [OpenAI's Pentagon Deal Closed in Hours. That Speed Is the Story.]() — Published 2026-04-01 OpenAI announced its classified Pentagon agreement on February 27, mere hours after the administration blacklisted Anthropic. Sam Altman later called the timing "opportunistic and sloppy." He was being generous with himself. The word he needed was "revealing." - **"Anthropic wins preliminary injunction in DOD fight as judge cites 'First Amendment retaliation'" — CNBC, March 26, 2026** — [Source]() — Published 2026-04-01 The most complete account of Judge Lin's ruling, covering the injunction's scope, the two-statute legal structure, and how the contract dispute actually started. - **"Judge blocks Pentagon's effort to punish Anthropic" — CNN, March 26, 2026** — [Source]() — Published 2026-04-01 Picks up what the CNBC piece doesn't lead with: the implementation delay, the Pentagon CTO's framing, and the due process findings that may matter more long-term. - **"Where We Stand" — Anthropic, March 2026** — [Source]() — Published 2026-04-01 Anthropic's own account of its red lines on autonomous weapons and mass surveillance — read it as advocacy, but it's indispensable primary sourcing on what the company says it refused and why. - **"What Everyone Is Missing About Anthropic and the Pentagon" — Internet Governance Project, March 8, 2026** — [Source]() — Published 2026-04-01 Published weeks before the ruling, this reframes the dispute as a governance vacuum rather than an ethics standoff — a structural reading the court's decision later bore out. - **"Towards a Science of AI Agent Reliability" — arXiv:2602.16666 / Fortune summary, March 24, 2026** — [Source]() — Published 2026-04-01 Princeton's framework for measuring agent reliability separately from capability — consistency, robustness, calibration, and safety each scored independently — and the uncomfortable gaps that appear when you do. ### [Foundations]() Conceptual clarity earned from building at scale - [Everything Grows When Nothing Has Failed Yet]() — By [Nora Kaplan]() · Published 2026-04-01 · Format: Before the Thinning Ninety-seven million monthly downloads. Every major lab shipping a framework. Thousands of MCP servers, most less than a year old. The landscape maps for AI agents have become so dense they're nearly unreadable. By every surface measure, the ecosystem is thriving. - [The Agent Toolchain Is Consolidating in Four Different Directions at Once]() — By [Rina Takahashi]() · Published 2026-04-01 · Format: Where Winners Emerge Playwright has 84,000 GitHub stars and roughly double the adoption of Selenium. By most measures, it won. So why did one of the most prominent AI browser tools just rip it out entirely and rewrite from scratch? Everyone talks about the agent toolchain consolidating. But the layers of the stack are moving at different speeds, in ways that contradict each other. The kind of movement you're watching tells you as much as which tool is ahead. - **120+ Agentic AI Tools Mapped Across 11 Categories** — [Source]() — Published 2026-04-02 The sheer count across eleven categories makes proliferation concrete. Useful less as a shopping list than as evidence of how fast category boundaries are blurring. - **MCP's 2026 Roadmap Makes Enterprise Readiness a Top Priority** — [Source]() — Published 2026-04-02 A close read of what "pre-RFC" actually means for enterprise features like audit trails and SSO-integrated auth. The gaps the protocol's own maintainers acknowledge tell you more than the features they ship. - **The Agentic Browser Landscape in 2026: A Complete Guide** — [Source]() — Published 2026-04-02 Browser tooling is where proliferation meets consolidation pressure first. This maps the bifurcation clearly: sites now serve two audiences, and the tooling reflects it. - **State of AI 2026 March Update: The Capabilities, Infrastructure, and Deployment Gaps** — [Source]() — Published 2026-04-02 Puts a number on the deployment gap: 83% of organizations planning agent initiatives, 29% with infrastructure to support them. That spread looks structural, not temporary. - **Why Model Context Protocol Is Suddenly on Every Executive Agenda** — [Source]() — Published 2026-04-02 Translates MCP's technical proliferation into the governance and attack-surface language that CIOs are actually using right now. Worth reading for the framing alone. - **Every Major AI Lab Ships Its Own** — [Source]() — Published 2026-04-02 LangChain holds 132,000 GitHub stars, but the real signal is elsewhere. OpenAI, Google, Anthropic, Microsoft, and HuggingFace each maintain their own orchestration SDK now. Over 15 active frameworks competing for how agents get built. A year ago, you could count the serious ones on one hand. - **MCP Hits 97 Million Monthly SDK Downloads** — [Source]() — Published 2026-04-02 Anthropic launched the Model Context Protocol in November 2024 at roughly 2 million monthly downloads. Eighteen months later: 97 million downloads, 10,000+ active public servers, every major AI provider on board, and a new home at the Linux Foundation. That growth curve is hard to overstate. - **Open-Source Browser Agents Top the Benchmarks Now** — [Source]() — Published 2026-04-02 Browser Use posts 89.1% on WebVoyager across 586 web tasks, leading the open-source pack. Ten-plus competing browser agent tools crowd the leaderboard behind it. The distance between open-source and proprietary shrinks with each release cycle. - **Watching Agents Work Became Its Own Industry** — [Source]() — Published 2026-04-02 Fifteen-plus dedicated observability platforms now exist for agentic systems. A category that barely had a name two years ago. LangSmith alone processes over 15 billion traces. The tooling covers distributed tracing, cost monitoring, failure-mode detection, and human-in-the-loop approvals. - **n8n Reaches 182k Stars as the Action Layer** — [Source]() — Published 2026-04-02 With 182,000 GitHub stars, 400+ integrations, and native LangChain support, n8n has quietly become the default no-code surface for AI agent workflows. A fair-code license and a 200,000-member community put it among GitHub's top 50 projects overall. - **GPT-5.4 Crosses the Human Baseline on Computer Use** — [Source]() — Published 2026-04-02 OpenAI's GPT-5.4 scores 75% on OSWorld-Verified, clearing the 72.4% human expert baseline for computer use. It arrives alongside Claude 4.6 Opus, Gemini 3.1 Pro, and NVIDIA Nemotron 3 Super in a March 2026 model wave that fuels everything listed above. - **83% Planning Agentic AI, 29% Ready for It** — [Source]() — Published 2026-04-02 HiddenLayer's 2026 survey of 250 IT and security leaders surfaces a 54-point gap between agentic AI ambition and operational readiness. Deloitte's data tells the same story from a different angle: ~70% piloting, ~30% prepared. That gap is where pilots go to stall. - **41% of Enterprises Are Still Just Watching** — [Source]() — Published 2026-04-02 Industry segmentation puts four in ten enterprises in the "Spectators" category: stated AI ambitions, negligible implementation. Fewer than one in five Spectators have adopted even basic enabling tools. Ambition without infrastructure is just a press release. - **MCP's Roadmap Openly Lists Its Missing Pieces** — [Source]() — Published 2026-04-02 The 2026 MCP roadmap names enterprise auth, audit trails, gateway behavior, and configuration portability as unresolved. A dedicated enterprise working group doesn't exist yet. At 97 million monthly downloads, these are production pain points, not theoretical shortfalls. - **Autonomous Agents Behind One in Eight AI Breaches** — [Source]() — Published 2026-04-02 HiddenLayer reports autonomous agents now drive more than one in eight AI-related breaches. Meanwhile, 5,800+ community-built MCP servers expand the attack surface with minimal gatekeeping. Worse: 31% of organizations can't say whether they've been breached at all. You can't govern what you can't see. - **The Bottleneck Moved From Models to Operations** — [Source]() — Published 2026-04-02 NVIDIA and Nutanix both announced infrastructure plays in March 2026, framing the constraint identically: the hard part is no longer building the model but managing infrastructure to securely run thousands of agents at scale. When hardware and platform vendors land on the same talking point independently, pay attention. - **$24B Testing Market Says Reliability Is the Constraint** — [Source]() — Published 2026-04-02 Fortune Business Insights values the automation testing market at $24.25 billion in 2026, projecting $84 billion by 2034. Services account for 55% of the total, which tells you something: reliability requires sustained operational investment, not a one-time tooling purchase. ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [The Agent's Alibi]() — By [Rina Takahashi]() · Published 2026-04-01 · Format: The Attribution Gap Seventy-eight percent of enterprises are piloting AI agents. Fourteen percent have made it to production. That gap has held steady through billions in improvement spending on data pipelines, prompts, tooling, and models. Every fix is genuine. The gap barely moves. - [One Production Failure, Thirty Scenarios, and the Debugging Infrastructure That Didn't Exist]() — By [Rina Takahashi]() · Published 2026-04-01 · Format: The Forensics Problem A customer support agent submits a credit card replacement before the customer agrees to it. The session log looks clean. Every tool call completed, every response was generated on time, and the conversation reached a natural close. Standard monitoring sees a successful run. The agent made a bad judgment call inside a reasonable-looking transcript, and nothing in the observability stack can tell you where. - [The Agent Incident Investigator Who Keeps a Second Set of Notes]() — Published 2026-04-01 *Practitioner's Corner sometimes talks to people who don't technically exist, but whose problems very much do. Declan Farrow is a composite: an "Agent Incident Investigator" assembled from the patterns, frustrations, and dark humor of practitioners who spend their days running post-mortems on production AI agent failures. His background in insurance claims adjustment is fictional. His observations are grounded in documented incidents and published research. His private log is imaginary. The things in it are not.* - [Agent Failure Attribution Is Entering Aviation's Systems Thinking Trap From the Opposite Direction]() — Published 2026-04-01 Aviation spent fifty years writing "pilot error" on accident reports before recognizing the phrase explained nothing fixable. The corrective was systems thinking. Failures became organizational, distributed, contextual. Better, and also convenient: when every failure belongs to the system, no specific component bears obligation to change. - **AI Didn't Break Your Production — Your Architecture Did** — [Source]() — Published 2026-04-01 Where you point the blame shapes what you fix. The attribution problem, stated plainly. - **The Multi-Agent Reality Check: 7 Failure Modes When Pilots Hit Production** — [Source]() — Published 2026-04-01 A useful anatomy of how multi-agent coordination degrades. The taxonomy reveals how many failure surfaces hide behind one symptom. - **AI Agents: Why the Gap Between Demo and Deployment Keeps Widening** — [Source]() — Published 2026-04-01 A counterintuitive argument: better tooling makes demos shinier, which makes the production gap harder to see. - **AI Agent Scaling Gap March 2026: Pilot to Production** — [Source]() — Published 2026-04-01 Survey of 650 enterprise leaders puts a number on the gap: 78% piloting, 14% in production. The drop-off is structural. - **How Veris AI and Lume Security Built a Self-Improving AI Agent with Microsoft Foundry** — [Source]() — Published 2026-04-01 A concrete case of turning production failures into reusable scenario families. The feedback loop, built deliberately. ### [Echoes]() Past infrastructure decisions echoing in today's production realities - [60 Hudson Street and the Persistence of Place]() — By [Rina Takahashi]() · Published 2026-04-01 · Format: The Inherited Topology There are pneumatic tubes running the full height of a building in lower Manhattan. They were installed in 1930 to carry paper. They're still in use, though not for paper. What they carry now, and why nobody ripped them out, turns out to be the same question. - [The Seabed Remembers: 170 Years on the Same Ocean Floor]() — Published 2026-04-01 In 1854, a U.S. Navy oceanographer named Matthew Fontaine Maury identified an undersea corridor between Ireland and Newfoundland. He called it the "Telegraph Plateau." The plateau turned out to be a myth. The corridor was not. - [A Conversation with a Data Center Site Scout Who Reads Railroad Maps for a Living]() — Published 2026-04-01 *Gail Parcell does not exist, which is a shame, because the data center industry could use her. She is a composite, assembled from the professional reality of infrastructure site selection consultants whose daily work involves reading geological and bureaucratic strata that most people never think about. We spoke with her over a video call from what she described as "a Holiday Inn in central Iowa that has better fiber connectivity than most office buildings in San Francisco." She did not explain how she'd tested this.* - [Why American Fiber Follows the Railroad]() — By [Rina Takahashi]() · Published 2026-04-01 · Format: Rights and Routes Overlay a map of America's long-haul fiber optic backbone onto its Class I railroad network and the two are nearly identical. Same cities, same corridors, same hubs. The engineering explanation is straightforward: cleared ground, existing infrastructure, accessible terrain. But the deeper pull involves legal permissions granted to railroad companies in the 1800s that never expired, never narrowed, and kept drawing each new generation of communication technology into paths laid down for steam locomotives. - [The Same Narrow Water]() — By [Rina Takahashi]() · Published 2026-04-01 · Format: Rights and Routes On land, old legal corridors pull new technology into old paths. At sea, geography does the locking. In February 2024, a drifting cargo ship dragged its anchor across the floor of the Red Sea and severed three submarine cables, knocking entire African nations offline. The route those cables followed through the Bab al-Mandab Strait was the same route telegraph cables followed in 1870. The strait hasn't widened. - **New York, New York! The NYC Carrier Hotel Microcosm** — [Source]() — Published 2026-04-01 Western Union's 1930 telegram hub at 60 Hudson still routes global internet traffic. The building outlasted the company. - **The Evolution of Carrier Hotels: 60 Hudson to Today** — [Source]() — Published 2026-04-01 Telecom deregulation met inherited real estate and created an industry. Good account of how carrier hotels actually happened. - **At the Bottom of the Sea: A Short History of Submarine Cables** — [Source]() — Published 2026-04-01 APNIC traces submarine cables from 1858 to fiber optics. The landing stations barely moved. The routes didn't either. - **I Traced Who Owns the Undersea Cables — The Map Is a Colonial One** — [Source]() — Published 2026-04-01 Colonial port cities became internet chokepoints. Big Tech now owns 70% of undersea bandwidth along those same routes. - **The Strategic Future of Subsea Cables: Egypt Case Study** — [Source]() — Published 2026-04-01 CSIS examines why so much global bandwidth squeezes through Egypt. Geography is policy by other means. - **U.S.-Iran War Threatens Gulf AI Infrastructure as Both Data Chokepoints Close** — [Source]() — Published 2026-04-01 Telegraph-era cable geography becomes live vulnerability as Red Sea and Hormuz disruptions hit simultaneously. History arriving as crisis. ### [Vision]() Where human-AI collaboration is heading - [The Infrastructure Nobody Budgeted For]() — By [Nora Kaplan]() · Published 2026-04-01 · Format: The Missing Rung A senior analyst pauses over a number that passed every automated check. Something about it, against a pattern she's absorbed across thousands of similar numbers over years, makes her stop. She couldn't fully articulate why. Nearby, a junior colleague registers the pause. Files it somewhere wordless. That transmission has no line item, no owner, no design document. Entry-level hiring in the U.S. is down 35% since 2023, and the savings show up immediately. What those roles were quietly producing has no corresponding line in the budget. - [IBM Bets That Watching Builds Knowing]() — By [Nora Kaplan]() · Published 2026-04-01 · Format: Two Bets on What Replaces It IBM is tripling entry-level U.S. hiring while deploying agents that handle the work those hires used to learn from. The redesigned junior role is oversight: reviewing agent output, flagging anomalies, collaborating across teams. It sounds like a better job. It might be. But the old path to expertise ran through years of doing things badly and recovering. If the new path runs through watching, does it arrive at the same place? - [PwC's Hourglass Org Runs on a Reservoir It Can't Refill]() — By [Nora Kaplan]() · Published 2026-04-01 · Format: Two Bets on What Replaces It PwC wants the agent-era workforce built around generalists: experienced professionals orchestrating AI across entire workflows instead of executing narrow tasks. The model works, for now, because today's generalists spent years as specialists first. They know what good looks like because they once produced it by hand. PwC's own workforce paper admits as much. So what happens at the top of the hourglass when the pipeline that filled it with depth stops running? - [Approximately Correct Since Day One — A Conversation From 2028]() — Published 2026-04-01 *This interview has not happened yet. It is set in April 2028, constructed from the trajectory of real shifts already underway — IBM's redesigned entry-level roles, KPMG's "Great Skills Reset" framework, the governance infrastructure being built as you read this. The subject is fictional. The situation he describes probably isn't.* - [When AI Knows Everything, Hiring Filters Shift From Knowledge to Cognition]() — Published 2026-04-01 McKinsey's Solve assessment doesn't quiz candidates on business frameworks. It watches them play. Over 110 gamified minutes, the tool tracks every click, every pivot, every data point examined or ignored. Process matters as much as outcome. - **How AI Is Changing the Nature of Entry-Level Work** — [Source]() — Published 2026-04-01 The 35% decline in entry-level postings, and why those roles were always doing double duty as paid education. - **The Bottom Rung Returns as AI Reshapes Entry-Level Jobs** — [Source]() — Published 2026-04-01 IBM's counterargument: triple entry-level hiring, but rebuild every role around judgment and oversight. The strongest case against the doom frame. - **AI and the Entry-Level Job** — [Source]() — Published 2026-04-01 HBR traces the pipeline forward: cut junior roles now, and in five years your senior talent pool is hollow. - **AI Shifts Expectations for Entry-Level Jobs** — [Source]() — Published 2026-04-01 Programmer employment dropped 27.5% while design-oriented roles held. New grads are now expected to slot in at a higher level from day one. ## More - [Publication guide]() - [Issue archive]()