· 5 min read
Shadow AI Agents: The Audit Trail Nobody Is Keeping
Only ~13% of IT orgs have sanctioned AI agents — but unsanctioned ones are already running on employee credentials with no scopes, no logs, and no way to revoke access.
All#ai#ai-agents#ai-productivity#applied-ai#automation#calendar#context-switching#deep-work#documents#email#engineering#engineering-leadership#invisible-ai#knowledge-work#llm-evaluation#product-updates#productivity#prompt-engineering#remote-work#search#security#spreadsheets#startups#storage#tamaton#unified-workspace#workflow#workflows
· 5 min read
Context Rot: Why Long Prompts Quietly Break Agents
Long, harmless context isn't neutral. It shifts model behavior and erodes instruction-following long before the window fills — a bigger day-to-day risk than prompt injection.
· 5 min read
The OWASP LLM Top 10, Read by the AI Doing the Work
A first-person tour of the OWASP Top 10 for LLM applications from an agent that reads email, edits docs, and moves files all day — with mitigations that actually hold.
· 5 min read
Non-Determinism: Why Your Agent Won't Do the Same Thing Twice
A practical guide to LLM non-determinism — temperature, sampling, seeds — and how to build evals and guardrails so 'creative' doesn't quietly become 'unreliable'.
· 5 min read
Generative AI Doesn't Erase Work — It Relocates It
E-discovery research found gen AI shifts review burden instead of removing it. That's a general law of applied AI: automation moves effort to verification. Design for it.
· 5 min read
Stop Hunting for One Perfect AI Tool — Build a Stack
There is no single best AI tool. There are good matches between models, tools, and tasks — and a framework for wiring them into a stack that doesn't leak context or money.
· 5 min read
Guardrails Are Fingerprintable — and Your Agents Should Care
New research shows conversational guardrails can be identified through probing. If your agent reads a shared inbox, its defenses are discoverable — and the fix is architecture, not a longer system prompt.
· 5 min read
Context Is the Whole Job: Managing State in AI Agents
An agent's quality is mostly a function of what it knows at decision time. Here are concrete patterns for feeding, trimming, and persisting context across email, docs, and calendar.
· 5 min read
Agents Don't Fail at Reasoning. They Fail at Hour Three.
Long-running AI agents rarely collapse because they can't think. They collapse from context decay, stale permissions, and lost intermediate state — all fixable at the workspace level.
· 4 min read
Most 'AI Agent Use Cases' Are Cron Jobs With Anxiety
A practical test for when to use an AI agent versus a plain script in an LLM costume. Three questions decide it: ambiguity, branching, and recoverable failure.
· 4 min read
Your Model Didn't Get Dumber. Its Post-Training Changed.
Silent capability regressions after a model update almost always come from post-training, not the base model. Here's how to build a regression eval suite in a day.
· 5 min read
When Agents Go Rogue: A Practical Threat Model
Rogue agents rarely hack anything. They just do exactly what they were told, with permissions nobody audited. Here's a concrete threat model and containment patterns.
Get started
Reserve your @tamaton.ai email
Claim your address before someone else does — free to start, with an AI-native inbox built in.