Anthropic reduces cache TTL in March 6th system downgrade

The AI tools your business depends on just got less predictable, and most people won't notice until it's too late. Anthropic quietly downgraded their cache time-to-live settings on March 6th, joining OpenAI in making stealth changes that affect how businesses use AI daily.

The Pattern Everyone's Missing

This isn't just about technical tweaks. We're seeing a clear pattern: AI companies making operational changes without proper communication to their business users. Anthropic's cache downgrade means Claude's responses might be less consistent between sessions, whilst OpenAI recently removed Study Mode from ChatGPT entirely—again, without warning.

The GitHub issue tracking Anthropic's change has generated significant discussion, but here's what matters for your business: these companies are treating enterprise users like beta testers, not customers with operational dependencies.

Why Stealth Changes Break Business Workflows

When you've built processes around AI tools, unexpected changes don't just cause inconvenience—they can break entire workflows. A cache downgrade affects response consistency, which matters if you're using Claude for customer service templates, content generation, or any repeatable business task.

We've seen this firsthand with clients who built automation around specific AI behaviours, only to find their systems producing different outputs after undocumented changes. One client's content approval process completely broke when their AI tool started generating variations of previously consistent outputs.

The broader issue is accountability. Traditional software vendors document changes, provide migration paths, and communicate with enterprise customers. AI companies seem to think they can operate like consumer apps whilst charging business rates.

What This Means If You Run a Business

First, your AI dependencies are more fragile than you think. Any business process that relies on consistent AI outputs is at risk of silent degradation. This isn't paranoia—it's risk management based on observable behaviour patterns.

Second, you need backup plans. The companies providing your AI tools don't consider your business continuity when making changes. They're optimising for their metrics, not your operations.

Third, this creates opportunities for businesses that plan properly. Whilst your competitors scramble with broken AI workflows, having robust fallback processes gives you a competitive advantage.

What To Do About It

1. 1.Audit your AI dependencies immediately. List every business process that uses AI tools, from content creation to customer service. Identify which ones would break if the AI output changed significantly.

1. 1.Create fallback procedures for critical processes. Don't rely solely on AI for mission-critical tasks. Have manual alternatives or multiple AI providers for essential workflows.

1. 1.Monitor AI tool performance actively. Set up simple tests to check if your AI tools are producing consistent outputs. A weekly spot-check can catch degradations before they affect customers.

1. 1.Document your AI workflows thoroughly. When changes happen, you need to know exactly what broke and how to fix it. Poor documentation turns minor AI changes into major business disruptions.

1. 1.Consider on-premises or hybrid solutions for critical applications. Self-hosted AI models don't change without your permission. For essential business processes, this stability might be worth the additional complexity.

The AI tools that seem indispensable today won't stay stable forever. The companies building them are still figuring out their business models, and your operational needs aren't their priority. Plan accordingly.