s01
Agent Loop
“One loop & Bash is all you need”
one tool + one loop = one agent
messageswhile Truetool_use
The Core Pattern - every coding agent is one loop
while True:
response = model(messages, tools)
if response.stop_reason != "tool_use":
return response.text
results = execute(response.tool_calls)
messages.append(results)Every coding agent is this loop. The model calls tools until it is done — everything else in these 20 lessons is refinement around it.
topFDEjobs
Forward Deployed Engineer roles
Building with Claude? Get paid for it.
Forward Deployed Engineers ship AI agents inside real customer systems. Open roles at Anthropic, OpenAI, Cognizant and more.
Up to $325k Anthropic, OpenAI, more Updated daily
Browse FDE jobs 20 of 20 lessons
Stage 1: Let the Agent Act
Core capabilities
s02
Tool Use
“Adding a tool means adding one handler”
the loop stays untouched; new tools register into the dispatch map
TOOL_HANDLERSdispatch mapconcurrency
s03
Permission System
“Set boundaries first, then grant freedom”
check what can run, what must stop, and what needs approval
PermissionRuleapproval pipeline
s04
Hook System
“Hook around the loop, never rewrite the loop”
add extension points without changing the main loop
PreToolUsePostToolUseextension points
Stage 2: Handle Complex Work
Plan, delegate, make room
s05
TodoWrite
“An agent without a plan drifts”
list the steps before starting; completion rate doubles
TodoItemplan-then-execute
s06
Subagent
fresh messages[]context isolation
s08
Context Compact
“Context always fills up -- have a way to make room”
four compaction steps reduce tool results first, then summarize history when it remains over the limit
tool_result_budgetsnip_compactmicro_compactcompact_history
Stage 3: Remember & Recover
Memory and resilience
s09
Memory System
“Remember what matters, forget what doesn't”
three subsystems: selection, extraction, consolidation
selectionextractionconsolidation
s10
Task System
“Big goals break into small tasks, ordered, persisted to disk”
a file-backed task graph that lays the groundwork for multi-agent coordination
TaskRecordblockedBydisk persistence
s11
Background Tasks
“Slow ops go background, agent keeps thinking”
background threads run commands; notifications inject on completion
threaded executionnotification queue
Stage 4: Run Long Tasks
Persist, background, schedule
s12
Cron Scheduler
“Fire on schedule, no human kick needed”
trigger tasks automatically by time
durable schedulingsession-scoped triggers
s13
Agent Teams
“Too big for one agent -- let teammates divide the work”
persistent teammates coordinate, claim ready tasks, and use task-bound working directories
persistent teammatesatomic task claimstask-bound worktreestyped protocols
s14
MCP Plugin
“Not enough capability? Plug in more via MCP”
connect external tools into the same tool pool
tool discoverynamespaced toolstool pool assembly
Stage 5: Coordinate Many Agents
Teams and protocols
s15
Integrated Harness
“Many mechanisms, one loop”
the mechanisms used by the integrated example share one harness
tools, runtime context, tasks, teams, scheduling, and MCP around one loop
s16
Workflow Runtime
“When the orchestration shape is fixed, put it in code”
saved workflows with resumable journals
script orchestrationlifecycle eventsjournal resume
s17
Goal Loop
“A goal decides when the loop may stop”
an independent evaluator reviews each proposed stop; impossible, failed, or over-limit goals return control to the user
goal gateconversation evaluationautomatic continuation
s18
s18
Stage 6: Extend & Assemble
Skills, MCP, the full loop
