Reference · AI Agents

The agent stack — one-page cheat sheet

AI agentsAgent loopMulti-agentAIAgents

Published

For the VP chair · print me · pairs with Lesson 1 & the glossary
The one sentence: an AI agent is a stateless text-prediction function, wrapped in a loop, given tools and a goal, run by ordinary code.

The two diagrams

REQUEST PATH            THE LOOP (the engine)
User                    ┌───────────────────┐
↓                     ▼                   │
Agent  (LLM+loop)    OBSERVE ─▶ THINK ─▶ ACT─┘
↓                  (result) (LLM picks)(run tool)
Tool   (model asks,        │
↓     runtime runs)      ▼  goal met? → DONE
Billing Service         else loop (to max-steps)
↓
Database

Definitions (compressed)

LLMStateless text→text fn. No memory, no hands.
ToolTyped fn the model may request (name+schema).
Tool callingModel emits request; runtime runs it.
AgentLLM + tools + loop + goal; model picks path.
Agent loopObserve→Think→Act→Repeat. Reconcile intent.
RuntimeApp server hosting loop; state, tools, limits.
MemoryWhat runtime reloads. RAM=context, disk=store.
WorkflowDev hardcodes path. Predictable, cheap.
Agent vs WFModel decides path. Flexible, costly.
MCPOpen standard to plug tools in. USB-C for AI.
MCP serverExposes tools/data (the device).
MCP clientConnects from host app (the port).
Tool registryCatalog advertised to model (svc discovery).
Single agentOne loop, one context. Monolith.
Multi-agentOrchestrator + workers. Microservices.

Mental models

LLM      = Brain (no memory, no hands)
Tool     = Hands / a callable function
Agent    = Brain + Tools + Loop + Goal
Runtime  = App server that runs the loop + rules
Memory   = What runtime reloads (RAM vs disk)
Workflow = Developer chooses the path
Agent    = Model chooses the path
MCP      = USB-C / JDBC for tools
(server=device, client=port)
Single   = Monolith    Multi = Microservices
Archetype= Same engine + diff tools + diff job

Archetype = same engine, diff job

Agent= EmployeeTools
CodingSW Engineerfiles, shell, tests, git
Exec-asstExec Assistantemail, calendar, browser
SupportSupport RepKB, tickets, account, refunds
SRESRE Engineermetrics, logs, deploy, rollback
BillingBilling Opsinvoices, payments, subs, refunds

Three rules for the VP chair

  1. Default to workflows; pick agents only when the path can’t be predetermined.
  2. Start single-agent; split to multi-agent like microservices — reluctantly, with evidence.
  3. To judge a pitch, ignore the brand: what tools? what loop? what guardrails?

Buzzword → engineering translation

BuzzwordWhat it actually is
LLM / foundation modelStateless text→text endpoint
Function calling / toolsTyped RPC model requests, you run
Agentic / autonomous AILLM in a while-loop + tools + goal
Reasoning / thinkingThe Think step — pick next action
Orchestration frameworkThe runtime (app server for the loop)
Context windowBounded RAM, re-sent every call
BuzzwordWhat it actually is
Memory / RAGExternal store + query injecting text
Workflow / chainHardcoded pipeline / DAG you wrote
MCPStandard tool interface; N×M → N+M
Multi-agent / swarmOrchestrator-worker microservices
Copilot / AI teammateAn archetype: engine + a role’s tools
HallucinationModel predicted plausible-but-wrong text

Ten memory rules (recite these)

  1. LLM — stateless text-prediction function; brain in a jar, no memory, no hands.
  2. Tool — a typed function the model requests but never runs; calling = model asks, runtime does.
  3. Agent — LLM + tools + loop + goal; the model picks the path.
  4. Loop — a K8s reconciliation loop for intent: observe, think, act, repeat.
  5. Runtime — the app server hosting the loop; holds state, runs tools, enforces guardrails.
  6. Memory — what the runtime reloads into the prompt; context is RAM, the store is disk.
  7. Workflow vs agent — you write the if/else vs the model writes it; choose by predictability.
  8. MCP — USB-C for tools; server = device, client = port, registry = catalog.
  9. Single vs multi — monolith vs microservices; split only when one context/role can’t cope.
  10. Archetypes — same engine + different tools + different job description.

5-minute review (weekly) & 30-minute deep dive (when stalled)

5-min: recall, don’t re-read — say the one sentence; draw both diagrams from memory; recite the ten rules; translate three buzzwords blind. 30-min: blank-page reconstruct all ten levels; then re-derive the chain of necessity (each layer forces the next: stateless LLM → tools → loop → runtime → memory → workflow/agent → MCP → single/multi → archetypes); then place one real product on the spectrum (workflow/agent? single/multi? which archetype? which tools?).