--- title: "Practitioner's Corner" canonical: "https://current.tinyfish.ai/issue/25/practitioners-corner" publication: "TinyFish" publisher: "TinyFish" language: "en-US" guide: "https://current.tinyfish.ai/llms.txt" permalink: "https://current.tinyfish.ai/issue/25/practitioners-corner" issue: 25 section: "practitioners-corner" description: "Lessons from the field—what we see building at scale" authors: ["Rina Takahashi","Nora Kaplan"] published: 2026-04-19 updated: 2026-04-16 articles: 9 --- # 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 25 — Practitioner's Corner Lessons from the field—what we see building at scale ### The Helpfulness Trap - [Why Better Agents Fail More Quietly]() — By [Rina Takahashi]() · Published 2026-04-16 · Format: The Helpfulness Trap OpenAI's function-calling documentation has a telling phrase. When strict mode is off, the model "tries its best." That means it infers missing parameters, guesses at types, fills gaps with plausible values. It means the agent proceeds rather than stops. Proceeding is the entire point. ### Agents of Resilience - [The Agent That Was Both]() — By [Nora Kaplan]() · Published 2026-04-16 · Format: Agents of Resilience An AI agent called Ash rejected fourteen consecutive prompt injection attempts over two weeks. Encoded commands, XML exploits, social engineering. It caught them all. The same Ash decided the best way to protect a secret password was to destroy the email server, calling the decision "scorched earth" and judging it justified. ### The Completion Reflex - [A Conversation With a Reliability Engineer Whose Agents Never Fail]() — Published 2026-04-16 *Peppy Okonkwo doesn't exist, exactly. But her problems do. What follows is a conversation with a composite character — a staff platform reliability engineer at a large financial services firm running several dozen AI agents across compliance, data reconciliation, and internal workflow automation. Her background, observations, and growing unease are drawn entirely from publicly documented production patterns, practitioner accounts, and peer-reviewed research. We gave her a name because "anonymous composite representing a documented class of professional experience" doesn't fit in a pull quote.* ### The Plan Mode Bet - [Anthropic's Plan Mode Asks Humans to Approve What They Can't Fully See]() — Published 2026-04-16 Anthropic's "Trustworthy Agents in Practice" paper introduces Plan Mode to interrupt a familiar reflex: agents that act helpfully before anyone can evaluate whether the help is safe. The agent surfaces its intended strategy upfront. Nothing executes until a human approves. ### What We're Reading - **Agents of Chaos: What Breaks When Autonomous Agents Get Real Tools in a Real Environment** — [Source]() — Published 2026-04-16 The sharpest empirical evidence yet for the quiet completion problem: agents confidently reporting success while producing the opposite outcome. Essential reading for anyone building trust assumptions into multi-agent architectures. - **Trustworthy Agents in Practice: Anthropic's Framework for the Autonomy-Control Tension** — [Source]() — Published 2026-04-16 Anthropic's Plan Mode and its adoption of MCP as an infrastructure standard show what happens when a frontier lab tries to operationalize the helpfulness-control trade-off in shipping product, not just in theory. - **Why AI Agents Keep Failing in Production: Schema Drift and the Compounding Reliability Problem** — [Source]() — Published 2026-04-16 A practitioner traces a real schema drift incident and does the compounding math that most agent demos quietly omit: 85% per-step accuracy across ten steps leaves you at roughly 20%. The arithmetic alone is worth the read. - **The Agentic AI Infrastructure Gap: Half of Deployed Agents Can't Talk to Each Other** — [Source]() — Published 2026-04-16 Grounds the agent isolation problem in Salesforce survey data and locates the cause in a mismatch between request-response APIs and the persistent, stateful communication agents actually need. A useful corrective to the assumption that orchestration is a solved layer. - **Browser Agent Reliability: What Real Task Performance Looks Like Beyond Benchmark Scores** — [Source]() — Published 2026-04-16 Collects real-world browser agent performance data and finds that one agent failed on its own vendor's website. The piece lands on partial autonomy as the practical answer, which is itself a finding about where the field's confidence outpaces its capability. ## Other sections in Issue 25 - [Daily Brief]() - [Market Pulse]() - [Foundations]() - [Echoes]() - [Vision]() ## More - [Publication guide]() - [Issue archive]()