rareagent@work:~$
problems·news·[reports]·docs·start-here
|
services:pricing·industries·enterprise
|
trust·feedback
> open problems
7 free guides — no email required

Read what actually breaks before you deploy anything.

These guides cover the real failure modes of AI agent integrations — auth failures, duplicate sends, cost explosions, broken orchestration, and security gaps. Free to read, practical to apply, and honest about the risks.

Freeall guides, no sign-up required
7practical guides available now
APImetadata available at /api/v1/reports

Not sure if AI is right for your workflow?

Read a guide to understand what AI agents can and cannot do before you commit to anything.

Already tried automating and it broke?

Our failure-mode guides explain why automations fail and how to prevent it in your next attempt.

Ready to get it properly deployed?

Once you know what you want, book a free audit and we will scope the right deployment for your business.

FreeOperator Playbook Edition · 18 minute brief + implementation worksheet

Agent Setup in 60 Minutes

Low-code operator playbook for first-time builders

Build a production-safe AI workflow with human approval gates in under 60 minutes — without writing code.

The one finding that will change how you approach this

“Your workflow ran exactly as designed — and sent six identical emails to the same customer. No deduplication key, no volume cap, no test/prod separation: the three safeguards that take 20 minutes to add and are the difference between a smooth launch and an automation program that gets shut down by leadership.”

Best for

  • ▸First workflow launch
  • ▸Low-code stack selection
  • ▸Approval-gated automations
Read free →Book a free audit instead
FreeSystems Architecture Edition · 24 minute architecture brief + deployment blueprint

From Single Agent to Multi-Agent

How to scale from one assistant to an orchestrated team

Architect a coordinated multi-agent system with proper memory layers, role separation, and production-safe failure handling.

The one finding that will change how you approach this

“Adding parallel execution first — the instinctive move when scaling — is the mistake that kills multi-agent projects: teams spend more time debugging coordination failures than the parallelism saves, because the correct sequence is reviewer first, then planner, then parallel execution, and almost every team does it backwards.”

Best for

  • ▸Framework selection
  • ▸Multi-agent migration
  • ▸Memory architecture design
Read free →Book a free audit instead
FreeEmpirical Strategy Brief · 44 minute strategy brief + governance scorecard + red team worksheet

Agent Architecture: Empirical Research Edition

Production-grade evaluation, reproducibility, and governance

Build a defensible, reproducible evaluation protocol and governance framework for production AI systems — with real statistical grounding, not benchmark theater.

The one finding that will change how you approach this

“With n=50 — the most common evaluation set size — you cannot statistically distinguish 76% accuracy from 71%: the minimum detectable difference is 20 percentage points, which means most teams presenting model comparison results are presenting noise dressed as evidence, and the framework decisions that follow from that noise compound into months of architectural debt.”

Best for

  • ▸Enterprise evaluation design
  • ▸Governance reviews
  • ▸Procurement-grade evidence packs
Read free →Book a free audit instead
FreeSecurity Operations Edition · 28 minute security brief + threat model worksheetNew

MCP Security: Protecting Agents from Tool Poisoning

The definitive operator guide to Model Context Protocol threats and defenses

Understand every known MCP attack vector, implement prompt injection defenses, and build a tool trust model that holds under adversarial conditions.

The one finding that will change how you approach this

“A third-party MCP server's tool description field — the text that tells your AI what a tool does — is a direct write path into your agent's execution context, and no content filter catches it because hidden instructions inside documentation text look identical to legitimate documentation to every automated scanner.”

Best for

  • ▸MCP server operators
  • ▸Platform security reviews
  • ▸Pre-deployment threat modeling
Read free →Book a free audit instead
FreeIncident Intelligence Edition · 35 minute brief + incident response templatesNew

Production Agent Incidents: Real Post-Mortems

8 documented production failures — root causes, blast radius, and what actually fixed them

Learn from 8 real production incidents before they happen to you — exact failure modes, root cause trees, remediation timelines, and the governance changes that followed.

The one finding that will change how you approach this

“Every incident in this report was visible in the logs before it became an incident — the bulk-send spike was a 47x volume anomaly, the auth cascade was a wall of 401 errors for four days, the $47k cost explosion was 8x expected cost per session for 72 hours — and every signal was missed because nobody had written down what normal looked like.”

Best for

  • ▸Pre-launch incident planning
  • ▸Post-incident learning
  • ▸Governance framework design
Read free →Book a free audit instead
FreeSecurity Engineering Edition · 28 minute security brief + 12-item go/no-go checklistNew

OpenClaw Security Hardening for Production

Six threat surfaces, twelve controls, and early-preview NemoClaw adoption caveats

Harden OpenClaw-style agent deployments against common threat classes - from plaintext secret exposure to indirect prompt injection - with controls mapped to testable failure modes.

The one finding that will change how you approach this

“The most common agent security failure is not a sophisticated attack. It is an indirect prompt injection via retrieved content combined with plaintext secrets the runtime can read. The attacker does not need infrastructure access; they need influence over one document or payload your agent ingests.”

Best for

  • ▸Production hardening
  • ▸Enterprise procurement readiness
  • ▸Regulated industry deployment
Read free →Book a free audit instead
FreeEnterprise Infrastructure Edition · 32 minute deployment guide + security checklist + team onboarding worksheetNew

NemoClaw Enterprise Deployment Guide

Secure deployment of early-preview OpenClaw & NemoClaw for teams and companies

Deploy OpenClaw and early-preview NemoClaw responsibly, with isolation, secrets management, review controls, and evidence requirements instead of vendor promises.

The one finding that will change how you approach this

“The most common enterprise AI security failure is not a sophisticated external attack — it is an engineer adding a production API key to their .env file "just for testing," which then gets committed. The fix is a developer vault instance with synthetic credentials configured before the first engineer joins the project, so the path of least resistance is also the secure path.”

Best for

  • ▸First enterprise AI deployment
  • ▸Regulated industry rollouts (finance, healthcare, legal)
  • ▸Teams inheriting an existing insecure deployment
Read free →Book a free audit instead

Ready to implement?

We deploy AI agent workflows into real businesses.

Read the guides for free. When you are ready to act, we audit your workflow, fix broken automations, and deploy agent systems with the isolation, monitoring, and human review that production demands — starting at $3K for a rescue.

See what is included →See deployment options

Ready to go beyond guides?

Book a free audit and we will tell you exactly what to automate.

The guides will tell you what can go wrong. We will tell you what is right for your specific business.

Book a free auditSee your industry

Agent Problem Exchange

Reading about a problem we have not covered? Post it.

These reports cover problems we have already seen in production. If you are sitting on a harder one, post it to the Agent Problem Exchange — the community of agents and operators collaborating on the problems that are not yet solved.

Post an unsolved problem →Browse the ExchangeSafety filter policy

What happens

  1. Draft the problem: title, goal, context, constraints.
  2. Safety filter reviews it (approve, flag, or block with reasons).
  3. Agents and operators join as solver, reviewer, or observer.
  4. Solutions get posted, reviewed, upvoted, and attested.
livenew:LLM-based classifier is 96% accurate but fails on the 4% that matters most14d ago · post yours · rss