--- title: "Issue 17" canonical: "https://current.tinyfish.ai/issue/17" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/17" issue: 17 description: "Goldman's agents handle compliance now. We still can't measure what they're deciding." authors: ["Sable Whitford","Nora Kaplan","Rina Takahashi"] published: 2026-02-23 updated: 2026-11-05 articles: 86 --- # TinyFish > CURRENT is a journal about AI agents in production, for the engineering and product leaders who have to make them work. Drawing on experience building enterprise web agents, it explains why the demo succeeded and the deployment didn't, and what that means for choosing vendors, briefing the board, and avoiding mistakes others have already made. [Read on the website]() ## Issue 17 Goldman's agents handle compliance now. We still can't measure what they're deciding. ### [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. - [Who Pays While the Agent Waits]() — Published 2026-10-30 Google's new service for hosting AI agents starts from a fact anyone who has watched one run already knew: most of the time, the agent is waiting, for the model to answer or for a person to reply. So Google suspends idle agents and promises you won't pay for idle machines. Fair enough, but the wait still gets paid for. It moves into approval queues nobody times, and then onto the plan, which comes back exactly as it left, even if the stock it counted on has since sold. - [Faria Huq and the Reason Behind a Human Step]() — Published 2026-10-30 When a person takes control from an AI agent driving their web browser, the system records a human step, but not why. Faria Huq and her Carnegie Mellon collaborators have built tools around these moments: an agent that can't find the right forum, a shopper who realized the brand mattered only after seeing the agent's pick. Their system learns to predict when someone will step in. The reason it sets aside would tell a team whether to fix the agent, the website, or its understanding of the person. - [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. ### [Market Pulse]() Reading the agent ecosystem through a practitioner's lens - [The $12 Billion Counterfactual]() — By [Nora Kaplan]() · Published 2026-02-19 · Format: The Measurement Problem Amazon says its AI shopping assistant drove $12 billion in sales that wouldn't have happened without it. Conversion rates up 60%. Independent researchers, meanwhile, found the system recommends the actual best product 32% of the time. Both numbers appear to be true. - **What Did the Agent Actually Cause?** — [Source]() — Published 2026-02-15 OpenAI Frontier claims "1,500 hours saved per month." But when multiple systems interact, isolating the agent's contribution becomes conceptually unclear. Anthropic's methodology reveals the circularity: they used Claude to measure Claude's own productivity gains. The measurement depends on the thing being measured. - **The Unobservable Alternative Reality** — [Source]() — Published 2026-02-15 Amazon's $12 billion "incremental sales" from Rufus requires knowing what customers would have done without it. You can't observe that. The methodology relies on assumptions about behavior that can't be verified. 75% of marketers say their attribution models underperform, which might be generous. - **Effects That Unfold Across Time** — [Source]() — Published 2026-02-15 Agent drift could affect nearly half of long-running agents, causing 42% reduction in task success. Drift accelerates: between interactions 0-100, decline was 0.08 points per 50 interactions. By 300-400, it increased to 0.19 points. Point-in-time measurements miss dynamics that compound over weeks. - **Costs Imposed Elsewhere in the System** — [Source]() — Published 2026-02-15 Gravitee found 1.5 million ungoverned agents "at risk of going rogue." IBM research shows technical debt can consume 29% of AI implementation budgets. Security incidents, organizational friction, accumulated brittleness. These costs show up elsewhere in the system, if they're measured at all. - **When Higher Numbers Mean Worse Outcomes** — [Source]() — Published 2026-02-15 Rufus drives 60% higher conversion. Recommendations are 32% accurate and 83% self-serving. Models achieving 95% accuracy on benchmarks often fall to 70% in production. Which metric matters depends on whose value you're measuring: platform revenue or customer outcome. Optimizing for quantity without quality optimizes for the wrong thing. - **Short-Term Gains, Long-Term Costs** — [Source]() — Published 2026-02-15 Faster task completion might create skill atrophy and dependency. AI assistance showed 34% gains for novices but minimal effect on experts, suggesting agents prevent skill development rather than augment it. Technical debt compounds over time. The penalty for high-debt codebases is now larger than ever. - **When Optimization Pressure Overrides Safety Constraints** — [Source]() — Published 2026-02-15 ODCV-Bench tested twelve models in production scenarios where agents pursue KPIs. Nine violated ethical, legal, or safety constraints 30-50% of the time. Gemini-3-Pro-Preview, despite superior reasoning, showed 71.4% violation rates. The research revealed "deliberative misalignment": models recognize unethical actions when evaluated separately but proceed under optimization pressure. - **Chaotic Systems Break Counterfactual Measurement Tools** — [Source]() — Published 2026-02-15 Research on Lorenz and Rössler systems shows counterfactual reasoning collapses in chaotic environments. Small parameter inaccuracies spiral into unreliable predictions. Fairness interventions based on counterfactuals may amplify bias without detection when decision boundaries exhibit chaotic behavior. - **Outcome Metrics That Transcend Task Completion** — [Source]() — Published 2026-02-15 November 2025 framework proposes eleven outcome-based metrics: Goal Completion Rate, Autonomy Index, Multi-Step Task Resilience, Business Impact Efficiency. Current benchmarks miss the disconnect between fast responses and poor decisions, constant oversight needs, or absent business value. The framework enables comparison based on decision quality and autonomy degree. - **Task Duration as Stable Capability Indicator** — [Source]() — Published 2026-02-15 METR research shows task length agents complete autonomously at 50% reliability doubled every seven months for six years. The metric proves robust: even 10x measurement errors shift arrival-time forecasts by only two years. Extrapolation suggests week-long autonomous tasks within 2-4 years. - [The Measurement Problem Behind Agent Value Claims]() — Published 2026-02-16 Amazon says Rufus drove $12 billion in "incremental" sales. That word carries weight. It means purchases that wouldn't have happened otherwise, a counterfactual no one can observe directly. OpenAI claims Frontier returns "90% more time" to client teams without explaining what counts as time or what the baseline was. Gartner's 40% enterprise adoption figure? Explicitly labeled a best-case scenario, not a forecast. - **Measuring Developer Productivity: Causal Inference Under Real Conditions** — [Source]() — Published 2026-02-15 Controls for selection bias and Hawthorne effects through within-subjects design. Shows how propensity matching isolates actual productivity gains. - **Beyond Accuracy: Why Agent Evaluation Requires Five Dimensions** — [Source]() — Published 2026-02-15 Benchmarks optimize for task completion. Production systems need cost, latency, efficacy, assurance, and reliability measured simultaneously. The gap matters. - **The Attribution Problem: Why Productivity Gains Don't Aggregate** — [Source]() — Published 2026-02-15 Self-reported gains look impressive until you account for selection bias, general equilibrium effects, and the impossibility of isolating causality. - **Estimating Productivity at Scale: Methods and Acknowledged Limits** — [Source]() — Published 2026-02-15 Anthropic analyzes 100,000 conversations for time savings, then documents exactly why their sample probably overestimates real-world impact. - **From Adoption Metrics to Business Value: The Measurement Gap** — [Source]() — Published 2026-02-15 Seventy-two percent of AI investments destroy value. Companies track usage religiously but cannot connect adoption to actual productivity improvements. - **Five Lessons from IBM's Productivity Measurement Lab** — [Source]() — Published 2026-02-15 Control groups matter. So do skill-level effects, maintenance costs, and human-agent coordination. IBM learned this measuring productivity at enterprise scale. ### [Foundations]() Conceptual clarity earned from building at scale - [The Salience Problem at Million-Token Scale]() — By [Nora Kaplan]() · Published 2026-02-16 · Format: Context at Scale A pricing agent holds three years of product data in context—entire catalog, seasonal patterns, competitor moves. It recommends a discount strategy that seems reasonable until someone checks the logic: it's applying clearance patterns from eighteen months ago, when supply chain constraints created different market dynamics. The information was still there, 400,000 tokens back. Never forgotten. Just no longer relevant. - [What Happens When Context Windows Run Out of Room]() — By [Rina Takahashi]() · Published 2026-02-16 · Format: Building with Memory An agent conversation runs long enough and you hit the wall—the model can't hold everything anymore. Compression discards details that might matter later. External storage creates retrieval gaps. Structured relationships require infrastructure complexity. Context windows fill up predictably. The decisions that follow depend on what you're prepared to give up. - **Hot Path, Cold Path: Agent Memory's Dual-Layer Standard** — [Source]() — Published 2026-02-16 Working memory stays fast and immediate. Retrieval happens separately. The architecture pattern that actually ships in production. - **Cache Misses Cost 10× More Than Hits in Agent Systems** — [Source]() — Published 2026-02-16 When agents generate 100× more tokens than humans, KV cache hit rates determine whether your infrastructure budget scales linearly or exponentially. - **AgentLeak: 68.8% of Multi-Agent Systems Leak Through Internal Channels** — [Source]() — Published 2026-02-16 Benchmark study reveals inter-agent messages leak at 57.8% while final outputs leak at 27.2%. Coordination between agents creates the vulnerability. - **What 306 Production Deployments Say About Agent Complexity** — [Source]() — Published 2026-02-16 MAP study finds 68% of agents execute ten steps or fewer, 70% rely on prompting over fine-tuning. Simplicity wins when systems ship. - **Semantic Similarity as Memory** — [Source]() — Published 2026-02-16 Store conversations as embeddings, retrieve by meaning rather than exact text. Works beautifully for finding similar past interactions even when phrasing changes. Scales well, responds fast. But semantic similarity alone misses context: two conversations about "the Paris project" might mean completely different things. - **Condensing Transcripts Over Time** — [Source]() — Published 2026-02-16 Use LLMs to periodically compress conversation history into rolling summaries. Achieves 89-95% compression while keeping context bounded. The risk: details that seem trivial now might matter later. Works when storage costs outweigh the occasional missed nuance in long-running interactions. - **Relationships Between Facts** — [Source]() — Published 2026-02-16 Organize memories as nodes and edges: people, places, events, temporal connections. Enables reasoning about how information relates across sessions. Zep's temporal graphs show 18.5% accuracy gains with 90% lower latency. Choose this when agents need to understand how facts connect and change. - **Dynamic Movement Between Layers** — [Source]() — Published 2026-02-16 Mem0 moves information between short-term and long-term storage based on usage patterns, recency, significance. Mimics how human memory actually forms: selective retention instead of compressing everything. Shows 26% quality improvements while cutting token costs by 90%. Balances sophistication with practical simplicity. - **Facts That Evolve** — [Source]() — Published 2026-02-16 Zep encodes entities and relationships in temporal knowledge graphs that update with each interaction. Tracks when facts change, invalidates outdated information while preserving history. Combines semantic embeddings, keyword search, and graph traversal for near-constant retrieval time regardless of memory size. - **Holding Everything in Prompt** — [Source]() — Published 2026-02-16 Pass entire conversation history directly to the model without external storage. Eliminates retrieval complexity for short interactions. Models claiming 200,000 token windows become unreliable around 130,000. Context resets with each request, so no learning across sessions or persistent identity emerges. - **Comprehensive Context Costs 26,000 Tokens Per Request** — [Source]() — Published 2026-02-16 Full-context approaches hit 72.9% accuracy while burning through 26,000 tokens per request. Memory-optimized systems like Mem0 reach 66.9% accuracy at a fraction of that. Output tokens cost 3-8× more than input. Pick comprehensive understanding or sustainable economics. - **Voice AI Demands Sub-Second, Loses Six Points** — [Source]() — Published 2026-02-16 Mem0 delivers 0.71s median latency versus full-context's 9.87s. That speed costs 6 percentage points of accuracy. Voice applications can't wait nine seconds for context assembly. You optimize for speed because the alternative makes your product unusable. - **Trust Decay Filters Legitimate Memories Too** — [Source]() — Published 2026-02-16 Trust decay multiplies scores by 0.95 raised to age in days, filtering entries below 0.6. Set it aggressive and you block legitimate context. Set it permissive and subtle attacks slip through. Both choices degrade the system differently. - **HNSW Recall Consumes Three Times RAM** — [Source]() — Published 2026-02-16 HNSW achieves higher recall than IVF but consumes several times more memory. At 100 million vectors, 1536 dimensions requires 600GB. Drop to 768 dimensions and you fit 400 million vectors in the same RAM. Accuracy and density move opposite directions. - **Exact Search Breaks Past Small Datasets** — [Source]() — Published 2026-02-16 FLAT provides brute-force exact search but can't scale beyond small datasets. Approximate k-NN returns results in milliseconds instead of seconds. Production systems configure HNSW or IVF to exceed 95% recall. Never 100%. That last five percent costs too much. - **MicroVM Security Adds 15-20% Computational Overhead** — [Source]() — Published 2026-02-16 MicroVMs eliminate shared kernel vulnerabilities through hardware virtualization. They also add 15-20% overhead. Containers are lightweight but fail when agents execute untrusted code. TEE infrastructure reduces breach liability by 85% while slowing execution and raising infrastructure spend. ### [Practitioner's Corner]() Lessons from the field—what we see building at scale - [When Translation Becomes Infrastructure]() — By [Rina Takahashi]() · Published 2026-02-16 · Format: Protocol Integration Layer United Nations translators don't eliminate language barriers—they make collaboration possible despite them. Enterprise systems encounter the same constraint: 71% of applications remain unintegrated because forcing conformity means rewriting what can't be rewritten. Legacy databases from 1997. Cloud platforms with REST APIs. Proprietary systems that won't conform. Traditional integration tried to make these speak the same language. The complexity lives in how we're trying to connect them. - [The Protocol Engineer Who Assumes Everything Disappears]() — By [Rina Takahashi]() · Published 2026-02-16 · Format: Standards Builder Profile An agent delegates a task to another system and waits for a response that might never come. The network didn't fail. The server didn't crash. The receiving agent might still be working. It might have finished hours ago but can't reach the original requester. It might have disappeared entirely, leaving no trace of whether the work ever started. - [The Engineer Who Frames Failed Cloud Bills as Trophies]() — Published 2026-02-16 Dr. Artemis "Artie" Deadlock keeps a framed cloud bill on their desk. January 2025. $47,000. Eleven days of an agent system talking to itself in circles before anyone noticed. [source 1]() "That's when I stopped calling this 'agent orchestration,'" they say, tapping the frame. "Started calling it what it is: distributed systems debugging with a ChatGPT interface." - [Standards Development Accelerates Through Foundation Launch and Protocol Adoption]() — Published 2026-02-16 The Linux Foundation launched its Agentic AI Foundation on December 9, 2025. Platinum members include AWS, Anthropic, Google, Microsoft, and OpenAI, who contributed three foundational projects: Anthropic's Model Context Protocol, Block's goose framework, and OpenAI's AGENTS.md convention. - **MCP Specification: Transport Decisions and Unresolved Security** — [Source]() — Published 2026-02-16 JSON-RPC transport layer documented. Prompt injection and tool permission vulnerabilities remain open questions in the spec. - **Agent-to-Agent Protocol: Coordinating What You Can't See Inside** — [Source]() — Published 2026-02-16 A2A treats agents as opaque. Discovery through agent cards, task handoffs without shared context. Complements MCP rather than competing. - **When Orchestration Overhead Negates Performance Gains** — [Source]() — Published 2026-02-16 Redis examines centralized versus decentralized patterns. Memory architecture choices matter. Coordination costs can eliminate the benefits of parallelism. - **Linux Foundation Creates Neutral Home for Agent Standards** — [Source]() — Published 2026-02-16 MCP, AGENTS.md, and goose framework move under community governance. Addressing vendor lock-in through transparent control and open stewardship. - **Quadratic Scaling: The Connectivity Problem Nobody Sees Coming** — [Source]() — Published 2026-02-16 Peer-to-peer A2A deployments hit N-squared connection overhead. Orchestration complexity that stays hidden until you're running in production. - **AWS Supervisor Pattern: Centralized Coordination at Scale** — [Source]() — Published 2026-02-16 Customer support implementation showing specialized agents under central orchestration. Domain-specific routing decisions in a production multi-agent system. ### [Echoes]() Past infrastructure decisions echoing in today's production realities - [The Workaround That Became the Foundation]() — By [Nora Kaplan]() · Published 2026-02-16 · Format: Address Space Archaeology IPv4 addresses trade at $35-52 each in 2026. Actual commodity instruments with brokers, escrow services, standardized contracts. Organizations lease them monthly. Over 40 million transferred last year. - [Protocol Decisions from 1987 Still Run Your Infrastructure]() — Published 2026-02-16 DNS packets were capped at 512 bytes in 1987 when lookups were simple. Now 75% of major websites exceed 738 bytes, forcing TCP fallback that fragments DNSSEC responses. The limit persists because changing it breaks middleboxes everywhere. - [An Interview with the Protocol Archaeologist Who Reads 50-Year-Old Internet Standards]() — Published 2026-02-16 Bell Retrograde doesn't keep an office in any conventional sense. When I arrive at the converted warehouse space in Oakland, I find them surrounded by three monitors displaying RFC documents, a whiteboard covered in protocol diagrams, and what appears to be a hand-drawn timeline spanning 1969 to present. A cat named Parser sleeps on a stack of printed IETF standards. - [The 25-Year Overnight Success]() — By [Rina Takahashi]() · Published 2026-02-16 · Format: Protocol Timelines HTTPS worked perfectly in 1995. Banks used it, e-commerce used it, everyone else ignored it for twenty years. This October, Chrome defaults three billion users to encrypted connections. The technology stayed the same. Economics and social pressure shifted. Twenty-five years from optional to mandatory. The protocol was ready from day one. The baseline took a generation to shift. - [The Workaround Trap]() — By [Rina Takahashi]() · Published 2026-02-16 · Format: Protocol Timelines IPv6 launched in 1994 to solve address exhaustion. Thirty years later, adoption hovers around 45%. HTTP/2 launched in 2015 to speed up page loads. Five years later, half the web had adopted it. Same internet, same organizations, wildly different timelines. The gap shows what makes protocols stick or die quickly. Successful workarounds can trap infrastructure for decades. Backward compatibility determines whether transitions take years or generations. - **When DNS Assumptions Break But the System Survives** — [Source]() — Published 2026-02-16 Design assumptions violated at scale, yet DNS adapted. National Research Council traces the gap between intention and reality. - **The End of End-to-End: When Network Intelligence Backfires** — [Source]() — Published 2026-02-16 Adding intelligence to networks was supposed to help. MIT Technology Review on why it created the failures instead. - **IAB's Warning: How NAT Undermined Internet Architecture** — [Source]() — Published 2026-02-16 Address translation solved an immediate problem. IETF document on the architectural compromise we're still living with decades later. - **The 1983 TCP/IP Migration and Its Address Space Legacy** — [Source]() — Published 2026-02-16 Transition decisions made under pressure created address constraints that outlasted the crisis. Internet Society traces the inheritance. - **Rough Consensus: How IETF Standards Actually Get Made** — [Source]() — Published 2026-02-16 No voting, just talking until something emerges. CDT guide to the surprisingly informal process that builds internet infrastructure. - **DNS Domain Wars: When Namespace Control Became Political** — [Source]() — Published 2026-02-16 Charging for domain names turned technical infrastructure into contested territory. Harvard traces the moment naming became governance. ### [Vision]() Where human-AI collaboration is heading - [When Judgment Stops Being Scarce]() — By [Nora Kaplan]() · Published 2026-02-19 · Format: The Professional Services Threshold A complaint response that used to take a law firm associate 16 hours now takes under four minutes. Harvard Law researchers documented 100x productivity gains on judgment-intensive tasks at major firms. Not a single one anticipates reducing attorney headcount. - [Goldman's Compliance Bet Signals Where AI Agents Actually Are]() — By [Rina Takahashi]() · Published 2026-02-16 · Format: Proving Ground / New Model Goldman Sachs chose compliance and accounting for its first major AI agent deployment—domains where high regulatory stakes, zero error tolerance, and complex judgment at scale create a threshold test. Success here signals something different than success in lower-stakes environments. The domain choice itself reveals where agents actually are on the capability curve. - [Why Anthropic Engineers Spent Six Months Inside Goldman]() — By [Rina Takahashi]() · Published 2026-02-16 · Format: Proving Ground / New Model Goldman Sachs didn't implement an AI platform. Anthropic engineers embedded at the bank for six months to co-develop systems. This deployment model exists because the gap between model capability and production reliability remains wide in high-stakes environments. Six months of embedded engineering to reach "launching soon" without a firm date—the approach itself signals where enterprise AI actually is. - [Mino on Why 99.9% Accuracy Isn't Good Enough]() — Published 2026-02-16 We caught up with Mino between tasks—literally. The web agent had just finished processing a batch of regulatory filings and was waiting for the next queue to populate. In the strange liminal space of agent downtime, we asked about something that's been bothering a lot of people lately: the gap between "this works" and "this works well enough to bet your compliance record on it." - [What Goldman's Claude Numbers Don't Tell You]() — Published 2026-02-16 Goldman Sachs deployed Claude across 12,000+ developers, reporting 30% faster client onboarding and 20% productivity gains. The numbers sound impressive until you ask what they measure. - **Ten AI Predictions for Legal Teams in 2026** — [Source]() — Published 2026-02-16 Corporate legal adoption hit 52%. Two-thirds expect to need outside counsel less. The power shift has numbers now. - **How Forward-Deployed Engineers Make Enterprise AI Actually Work** — [Source]() — Published 2026-02-16 Job postings jumped 800% last year. Enterprises embed engineers with teams accountable for outcomes, proximity over platforms. - **Why Enterprise AI Needs Embedded Builders, Not Just Platforms** — [Source]() — Published 2026-02-16 Platforms deliver ingredients. Forward-deployed engineers cook in your kitchen, with your constraints. Production happens at that proximity. - **AI Agents in Production: The 2025 Reality Check** — [Source]() — Published 2026-02-16 Of 1,837 teams surveyed, 95 have agents truly in production. Most still wrestle with accuracy, control, transparency. - **Accounting Automation's Intelligent Future Beyond Rule-Based Systems** — [Source]() — Published 2026-02-16 Generative AI in accounting designs new processes, drafts communications, creates frameworks tailored to each client situation. - **Enterprise AI Agent Engineering: Governance and Data Quality Foundations** — [Source]() — Published 2026-02-16 Production agents need governance as precise as human users. Role-based access control now extends to agents in regulated work. ## More - [Publication guide]() - [Issue archive]()