Insights on enterprise AI
Practical guidance, case studies, and analysis on AI strategy, RAG, agents, automation, and integration — written by IDS engineers and consultants.
The token economics of scale: keeping AI costs flat as usage 10×s
Token cost grows linearly with usage. Five well-known levers — model routing, prompt caching, response budgets, batch APIs, eval-driven downgrades — compound to flatten that curve. Most teams pull them out of order. The eval suite is the prerequisite for the biggest savings.
Beyond chatbots: agentic AI is finally crossing into core enterprise workflows
Agentic AI — models that plan, call tools, verify their own outputs — has crossed the threshold from demo to production. Three things change in the architecture, three workflows earn it first, and one rule of thumb tells you when not to reach for an agent.
Building an LLM threat model: a 7-step framework for enterprise AI
STRIDE doesn’t fit. OWASP’s LLM Top 10 is a taxonomy, not a process. Compliance checklists ask the right questions for the wrong systems. A seven-step framework that produces a CISO-signable artifact and a runbook your engineering team will actually use.
Beyond prompt injection: data exfiltration risks in enterprise AI agents
Prompt injection is the entry point. The interesting question is what the agent does next. Four exfiltration patterns appear repeatedly in real enterprise AI agent deployments — each one has an architectural remediation, not a prompt-level one.
Five prompt injection patterns most security teams aren't testing for
Direct injection is the easy one. The four patterns that get past production red-teams — indirect injection via retrieved documents, tool-call hijacking, multi-turn context manipulation, encoding tricks — are the ones worth running before you ship.
Showing 5 of 15 articles.