Share

Field Dispatch

The Question Compliance Can't Answer

By Nora Kaplan— December 25, 2025

Feature image for article: The Question Compliance Can't Answer

"Show me why the agent flagged this price as anomalous."

The compliance officer reviewing our web agent deployment had the output—price flagged, alert generated. She had error rates: 99.2% accuracy. She had logs: timestamp, site, data, decision. What she didn't have: the reasoning chain. Which of 47 data points triggered the flag? How did conflicting regional information get weighted? The agent worked. She couldn't evaluate it. Compliance frameworks assume deterministic software. Agents operate probabilistically. Same input, different output depending on what the model learned yesterday.

Field Dispatch

The Question Compliance Can't Answer

"Show me why the agent flagged this price as anomalous."

The compliance officer reviewing our web agent deployment had the output—price flagged, alert generated. She had error rates: 99.2% accuracy. She had logs: timestamp, site, data, decision. What she didn't have: the reasoning chain. Which of 47 data points triggered the flag? How did conflicting regional information get weighted? The agent worked. She couldn't evaluate it. Compliance frameworks assume deterministic software. Agents operate probabilistically. Same input, different output depending on what the model learned yesterday.

Nora Kaplan

Nora Kaplan, former collaboration platform product leader turned technology writer. Studied human-computer interaction and spent years designing tools for knowledge work. Now writes about AI agents, work transformation, and how enterprise software reshapes human capability at TinyFish.

Builder Profiles

The Browser Engineer Whose Work Becomes Your Automation Challenge

By Rina Takahashi— December 24, 2025

Feature image for article: The Browser Engineer Whose Work Becomes Your Automation Challenge

A browser skips rendering off-screen content—faster page loads, satisfied users. But for web agents running thousands of concurrent sessions, elements that exist in the DOM yet won't render until scrolled into view? That's operational complexity disguised as optimization. The performance win that delights consumers creates unpredictability for automation. These tensions don't emerge randomly. They're engineered into browser behavior by people making careful trade-offs about security and speed—trade-offs that become the infrastructure challenges web agents must solve at scale.

Builder Profiles

The Browser Engineer Whose Work Becomes Your Automation Challenge

By Rina Takahashi— December 24, 2025

A browser skips rendering off-screen content—faster page loads, satisfied users. But for web agents running thousands of concurrent sessions, elements that exist in the DOM yet won't render until scrolled into view? That's operational complexity disguised as optimization. The performance win that delights consumers creates unpredictability for automation. These tensions don't emerge randomly. They're engineered into browser behavior by people making careful trade-offs about security and speed—trade-offs that become the infrastructure challenges web agents must solve at scale.

The Number That Matters

15% of API Changes Break Compatibility. 3% Actually Matter.

A study of 317 Java libraries across 9,000 releases found that 14.78% of API changes break backward compatibility. Less than 3% of those breaking changes actually impact client applications in production.

The gap exists because developers introduce breaking changes for new features and technical debt, not because existing clients need them. Methods take 59% of the hits, types 36%, fields just 5%. Most changes never trigger failures.

Teams monitor constantly. The work is figuring out which 3% will cascade through your stack before they do.

The Number That Matters

15% of API Changes Break Compatibility. 3% Actually Matter.

A study of 317 Java libraries across 9,000 releases found that 14.78% of API changes break backward compatibility. Less than 3% of those breaking changes actually impact client applications in production.

The gap exists because developers introduce breaking changes for new features and technical debt, not because existing clients need them. Methods take 59% of the hits, types 36%, fields just 5%. Most changes never trigger failures.

Teams monitor constantly. The work is figuring out which 3% will cascade through your stack before they do.

TAKE NOTE

Research methodology: Analyzed 260,000 client applications using APIDIFF tool across daily commits, detecting 17 types of breaking changes in popular libraries over multiple years.

Production costs: 67% of developers report unexpected API breaking changes, with enterprise downtime averaging $5,600 per minute when changes hit production systems undetected.

Temporal acceleration: Breaking change frequency increases as libraries mature, widening the gap between theoretical compatibility risks and actual production failures over time.

Versioning overhead: Point-to-point versioning with four API versions costs 45% more than compatible versioning, requiring independent development, patching, and deployment across versions.

Cascade example: Major news API silently changed response format, breaking 200+ applications within hours across financial trading bots and research platforms simultaneously.

Field Notes from the Ecosystem

December brought infrastructure changes that expose how platforms are rethinking resource allocation. Rate limiting is evolving from simple request counting to token budgets and complexity-based pricing. Three major platforms shifted their models this month alone.

The gap between agent system hype and production reality persists. The most successful implementations use simple orchestrators calling existing APIs, not complex frameworks. Browser automation libraries now bypass sophisticated defenses at 95% success rates. Compromised API keys without rate limiting become unlimited attack vectors. The infrastructure keeping the web operational grows more complex by the month.

Field Notes from the Ecosystem

December brought infrastructure changes that expose how platforms are rethinking resource allocation. Rate limiting is evolving from simple request counting to token budgets and complexity-based pricing. Three major platforms shifted their models this month alone.

The gap between agent system hype and production reality persists. The most successful implementations use simple orchestrators calling existing APIs, not complex frameworks. Browser automation libraries now bypass sophisticated defenses at 95% success rates. Compromised API keys without rate limiting become unlimited attack vectors. The infrastructure keeping the web operational grows more complex by the month.

Rate Limiting

Pipedrive Switches to Token Budgets

December 2 rollout for all new signups. Daily budget: 30,000 base tokens × plan multiplier × seats. Endpoints consume tokens by complexity and resource demand. Burst limits run on rolling 2-second windows per user.

API Deprecation

Instagram Basic Display Reaches End-of-Life

Shutdown completed December 4. Migration to Graph API mandatory. New constraint: 200 calls per hour per Instagram account. Ten connected accounts yield 2,000 hourly requests total, each maintaining isolated rate limit pools.

Complexity Pricing

Atlassian Moves to Points-Based Limits

Enforcement begins February 2, 2026 for Jira and Confluence Cloud. Calls consume points based on work performed: data volume, operation complexity. Base cost 1 point per request plus per-object charges. Cap: 500,000 points hourly.

Agent Architecture

Enterprise Agents Stay Thin and Simple

Production deployments use LLMs as thin orchestrators calling existing APIs as tools. The shift: from clicking buttons to saying "generate report" while pipelines handle execution. Simple, composable patterns outperform complex frameworks.