Multi-agent systems that actually ship inside a regulated enterprise: orchestration, tool use, guardrails, and the operating model around them. Grounded in CRYO (adopted by every core Optum line of business) and Snowboard (3,200 bots at 99.7% uptime).
Agentic AI in production · 5 min read · with a concept loop
Durable execution replays a workflow's own history, not the actions it takes outside itself. The effect ledger framework, and what leaders should fund.
workflow orchestrationreliabilityenterprise architectureagent runtime
Agentic AI in production · 5 min read · with a concept loop
A tool call that returns success is not proof an action happened. The four-gate chain shows what AI agents must verify before it counts.
Agentic AI in production · 5 min read · with a concept loop
Multi-agent AI rarely beats one accountable decision process. The sealed-evidence arbiter shows when coordination earns its cost and where risk hides.
multi-agent systemsagent governancerisk managementcost and unit economics
Agentic AI in production · 5 min read · with a concept loop
Agent skills are mixed evidence, not a proven lift. The three-gate skill contract separates discovery, execution and outcome so leaders fund what works.
agent skillsagent governancetool use and integrationevaluation and benchmarks
Agentic AI in production · 5 min read · with a concept loop
Self-improving AI agents need a sealed acceptance loop that keeps the grader, the test data and rollback authority out of the proposer's reach.
self improving systemsagent governanceautomated gradingevaluation and benchmarks
Agentic AI in production · 5 min read · with a concept loop
More reasoning tokens do not guarantee a better answer. The four-bottleneck model shows where to spend inference compute, and where to stop.
planning and reasoningcost and unit economicsinference servingwhat to fund
Agentic AI in production · 5 min read · with a concept loop
Multimodal AI succeeds by preserving where each fact came from. The evidence ledger framework for document, voice and video pipelines leaders can trust.
multimodal AIretrieval and groundingdata provenancestructured outputs
Agentic AI in production · 5 min read · with a concept loop
Multi-agent AI adds coordination cost and shared mistakes. The delegation contract shows when splitting work across agents actually pays off.
multi-agent systemsagent governancecost and unit economicssecurity
Agentic AI in production · 5 min read · with a concept loop
AI agent loops need a controller that can accept, repair, wait, reconcile, escalate or stop, not one that runs until it claims success.
agent runtimeagent governancereliabilitycost and unit economics
Agentic AI in production · 5 min read · with a concept loop
Before AI agents share raw internal state, prove transfer survives a scrambled message, compatibility holds across models, and confidentiality is real.
multi-agent systemsmemory and statesecurityprivacy and access control
Agentic AI in production · 5 min read · with a concept loop
Most agent failures start with the wrong amount of runtime. The runtime ladder matches five levels of infrastructure to what each task needs.
agent runtimeworkflow orchestrationsecurityvendor and build decisions
Agentic AI in production · 5 min read · with a concept loop
A realistic robot simulation is not proof a plan is safe. The four-layer envelope shows where independent verification has to sit.
simulation and world modelsmultimodal AIsafety and red teamingagent governance
Agentic AI in production · 5 min read · with a concept loop
Browser and computer-use agents need proof, not a success toast. The three-proof model for grounding, authority and verified effect.
Agentic AI in production · 5 min read · with a concept loop
Passing tests and large diffs are not proof of good code. The four-gate proof loop shows what to fund, measure and protect when agents patch repositories.
coding agentsagent governancesecurityevaluation and benchmarks
Agentic AI in production · 4 min read · with a concept loop
Agentic research agents can find sources and rerun code well. Only external checks turn a claim into validated science. The claim ladder makes that decision.
research due diligenceevaluation and benchmarksformal verificationaudit trails
Agentic AI in production · 3 min read · with a concept loop
Rule-based RPA breaks. Agents reason, plan and adapt. How we replaced an eight-figure RPA contract with 3,200 in-house bots at 99.7% uptime.
multi-agent systemsvendor and build decisionsROI and value measurementproduct and workflow design
Other topics
