CONCEPT Cited by 1 source

Agentic AI infrastructure challenges (eight-axis checklist)¶

Definition¶

Tyler Akidau's eight-axis checklist of what an enterprise has to solve to deploy agentic AI safely. Each axis names a concrete infrastructure problem that agent deployment surfaces. Together they form a canonical shopping list for evaluating "what does my agent platform need?" — and for classifying which vendor / substrate / pattern answers which axis.

Canonical source: Akidau's How to safely deploy agentic AI in the enterprise talk recap.

The eight axes¶

#	Axis	One-line meaning
1	Context building and maintenance	Keep the agent's knowledge base / vector store / analytical tables fresh with business data.
2	Context querying	Retrieve the right context at the right time (RAG, semantic search, SQL).
3	Authentication	Agent has access only to the data it's supposed to; can't escalate.
4	Governance	Enforce policies (PII, regulatory) uniformly across the agent fleet.
5	Auditing	Know what every agent did — full inputs + outputs, not just metadata.
6	Replay and validation	Re-run past interactions to confirm agents behave correctly.
7	Routing	Send the right subset of data to the right agent (or decide not to use an agent at all).
8	Multi-agent coordination	Synchronise multiple agents in complex workflows.

Verbatim:

"If you want to deploy agentic AI safely and effectively, you need to prioritize several non-trivial moving pieces: context building and maintenance, context querying, authentication, governance, auditing, replay and validation, routing, multi-agent coordination."

Akidau's load-bearing claim: six of eight are data-streaming problems (the other two — context querying + authentication — fall outside streaming's scope).

The streaming reduction¶

The six streaming-reducible axes map to existing wiki primitives:

Axis	Wiki primitive
1. Context building/maintenance	CDC + ETL into vector / OLAP stores; patterns/cdc-fanout-single-stream-to-many-consumers
4. Governance	Enforcement at interconnection points — the agentic data plane / MCP proxy
5. Auditing	patterns/durable-event-log-as-agent-audit-envelope — full inputs + outputs captured as first-class events
6. Replay + validation	patterns/snapshot-replay-agent-evaluation — replay audit log to validate agent behaviour
7. Routing	patterns/dynamic-routing-llm-selective-use — use AI where it wins, route to ML/heuristics otherwise
8. Multi-agent coordination	patterns/multi-agent-streaming-coordination — streaming broker as decoupled coordination substrate

The remaining two:

Context querying — vector DB + semantic search + SQL are the answer shapes. RAG covers the inference-time retrieval axis; a multi-modal catalog covers the discovery axis. Streaming doesn't serve this.
Authentication — IdP / OIDC / OBO substrates serve this; see concepts/short-lived-credential-auth + OBO authorization. Streaming doesn't serve this either.

Why this checklist matters¶

Evaluation rubric for agent platforms. When evaluating a candidate agent platform (Databricks Unity AI Gateway, AWS Bedrock Agents, Anthropic MCP, Redpanda ADP, Snowflake Cortex), the eight-axis checklist lets you ask "which of these eight does this product solve, and which does it outsource?" A platform claiming to be the "complete agentic data plane" should answer all eight credibly; platforms that solve 2-3 axes well and require partners for the others should be evaluated on the integration seams.

Gap enumeration for in-house builds. For teams building their own agent infrastructure, the checklist enumerates what must be solved before production deployment. Each unaddressed axis is a risk vector: unaudited agents can't be debugged; ungoverned agents can leak PII; unreplay-able agents can't be validated against regressions.

Shared vocabulary for cross-team discussions. The checklist gives platform teams, security teams, compliance teams, and application teams the same eight names for the same eight problems — reducing the coordination overhead of multi-team agent deployments.

Relationship to other framings on the wiki¶

concepts/governed-agent-data-access — Gallego's 2025-10-28 two-axis framing (access controls + observability) is a compression of axes 3 (authentication), 4 (governance), 5 (auditing) on this checklist. Gallego's framing is more mechanism-load-bearing; Akidau's eight-axis is the full-shopping-list altitude above it.
concepts/autonomy-enterprise-agents — the capability-side framing; the eight-axis checklist is what's required to deploy autonomy safely.
concepts/agent-dnd-alignment-framing — the eight axes together are the governance + auditing infrastructure that moves agents leftward from chaotic-good default toward lawful-good operable.
concepts/streaming-as-agile-data-platform-backbone — the structural claim this checklist rides on; six of eight axes reduce to streaming problems.

Caveats¶

Editorial decomposition, not mechanism spec. Each axis has deep mechanism complexity (policy engines for governance, replay determinism for LLMs, consensus semantics for multi- agent coordination) that the checklist doesn't engage.
Vendor-position framing. Akidau is Redpanda's CTO; the "six of eight are streaming problems" framing is structurally aligned with Redpanda's product positioning. Other vendors would decompose differently (Databricks would route more axes through the Catalog + Gateway; AWS would route more through Bedrock + IAM; Snowflake through Horizon + Cortex).
Axis coupling ignored. The eight axes are presented as independent; in practice they compose — e.g. governance + auditing share the same durable log substrate; replay + validation require both.
No axis priority. The list is unordered; in production, authentication + governance are prerequisites for everything else (an unauthenticated ungoverned agent shouldn't reach context-building).
No capability-axis. The checklist covers safety infrastructure; orthogonal capability infrastructure (tool surface breadth, model selection, context window management, agent-loop orchestration) isn't engaged.

Seen in¶

sources/2026-02-10-redpanda-how-to-safely-deploy-agentic-ai-in-the-enterprise — canonical wiki introduction. Akidau uses the eight-axis checklist as the core argument of his Modern Data Infrastructure Summit talk, pitching streaming as the substrate that answers six of the eight.