Training · ComposerMoveemergingverified

Agent-Builder Dojo

also known as dojo, AI builder dojo, agent dojo, intensive build sprint, agentic AI bootcamp

A short, intensive, small-group practice den lasting days to weeks. Builders repeatedly build real agents against real business problems under close facilitation. Measurable output — typically time-to-deploy or hours-saved — is the exit condition.

How the learner advances

flowchart TD A[Cohort of builders with real business problems] --> B[Problem scoping session narrow to shippable scope] B --> C[Build-deploy-review cycle runnable artefact every cycle] C --> D[Facilitated unblocking architecture / prompts / tools] D --> E{Agent deployed outside notebook?} E -- No --> C E -- Yes --> F[Capstone demo live agent + measured outcome] F --> G[More capable builder + production-candidate agent]

Intent. Ship at least one production-candidate agent per participant in a compressed, high-accountability build environment where the facilitator unblocks rather than lectures.

When to apply. Use this move when a team of builders already has foundational pattern knowledge and needs hands-on repetition against their own real problems — not when participants still need concept introduction. A dojo is the right follow-on to a build course; it is the wrong choice as a first step for complete beginners.

Threshold — earns the next step. Each participant has shipped at least one production-candidate agent with a documented, measured outcome by the final session.

Masterpiece — the artifact that proves it. A running, deployed agent built against a real business problem from the participant's own organisation, with a measured outcome — tasks automated, hours freed, or quality metric improved — demonstrated live at the capstone session.

Example scenario

A financial services firm runs a 4-week internal Agent-Builder Dojo with 12 engineers who completed the company's agent build course six weeks earlier. Each engineer brings a real problem from their team. One engineer is targeting the manual process of reconciling daily trade confirmations — currently two hours of analyst time per day. In the scoping session, the facilitator helps her narrow the problem to a single agent that reads incoming confirmation PDFs, extracts key fields, and flags discrepancies against the internal ledger. By end of week one she has a local prototype. The facilitator unblocks a tool-plumbing issue where the PDF parser was returning inconsistent JSON — a 20-minute fix that would have cost her two days alone. By week two the agent is deployed in a test environment reading real PDFs. By week three it is in production on a 10-trade daily sample. At the capstone demo in week four she shows the agent running live against that day's confirmations: 47 trades reconciled in 4 minutes, two discrepancies flagged, zero analyst time. Measured outcome: 1.8 analyst hours saved per day. The masterpiece is the running reconciliation agent with a documented time saving — not a notebook, not a demo script, but a deployed system processing real trade data.

Facets

Container — dojo
Mode — hands-on-buildbyo-problemmentored-unblockcapstone
Reach — team
Persona — builderanalyst-ops
Craft (AI Fluency) — delegationdescriptiondiligence

Inputs

Cohort of 6–20 builders with shared technical level — Participants who have at least basic familiarity with agent concepts and Python; ideally they have completed a build course or equivalent self-study. The shared technical baseline prevents the facilitator from splitting attention across wildly different starting points.
Real business problems from participants' own organisations — Each participant brings a specific, scoped problem from their own work context — not a toy problem. Solving real problems is what produces measurable outcomes and prevents the dojo from becoming academic.
Facilitator with agent-architecture depth — A practitioner who can unblock on architecture decisions, prompt structure, tool plumbing, and framework selection. The facilitator's role is reactive unblocking, not proactive lecturing.
Build-deploy infrastructure — Cloud access, API keys, and deployment tooling so participants can push working agents by end of each cycle, not just run them locally in notebooks.

Outputs

A more capable learner — A builder who has shipped an agent under real constraints and can describe the design decisions, failure modes encountered, and the next iteration needed.
Masterpiece: a production-candidate deployed agent — A running agent — built by the participant against their own real problem — with a measured outcome such as tasks automated, hours freed, or error rate reduced.
Aggregate outcome report — A post-cohort summary tracking time savings, task-automation counts, or other business-impact metrics across all participants — the evidence base for continuing investment in the dojo programme.

Steps (4)

Problem scoping session
Each participant presents their real business problem to the group. The facilitator and peers help scope it to the minimum agent design that can be shipped within the dojo's time box. Problems that are too broad get narrowed; problems that are trivially solvable without agents are redirected.
Build-deploy-review cycles
Daily or weekly cycles in which each participant builds, deploys something runnable, and gets a short review from the facilitator and peers. The cycle rhythm creates accountability and surfaces blockers early. No lecture-only sessions are permitted; every session produces a runnable artefact.
Facilitated unblocking
When a participant is stuck — on architecture, prompts, tool integration, or deployment — the facilitator unblocks in real time. Unblocking sessions are public when the problem is likely shared; private when it is specific to one participant's stack.
Capstone demo and measurement
Each participant presents their shipped agent with a measured outcome. The demo is live — the agent runs during the presentation. Outcomes are recorded in aggregate. Participants who have not shipped are not considered to have completed the dojo.

Principles

Real problems only — a dojo that allows toy problems produces toy skills.
Facilitation is unblocking, not teaching — any session that turns into a lecture has lost the dojo format.
Ship something runnable every cycle — an agent that only runs in a notebook has not been built.
Measure what changed — time saved, tasks automated, error rate reduced: the dojo's proof is a number, not a feeling.

Unlocks methodologies (3)

A learner who completes this pattern is equipped to execute these methodology families:

Agent Construction Multi-Agent Design Coordination

Known uses (4)

Known failure modes (3)

[toy-problem-drift]
The anti-pattern of participants scoping down to problems that are safe and easy rather than real and hard. When the dojo permits toy problems, participants ship agents that do not survive contact with production and the measured outcome is zero.
[lecture-creep]
The anti-pattern of the facilitator filling dead time with concept explanations. Once a session turns into a lecture, the dojo has become a classroom and the hands-on accountability structure collapses.
[notebook-only-deployment]
The anti-pattern of calling a Jupyter notebook that runs locally a 'deployed agent'. If the agent cannot be invoked by someone other than the builder without help, it has not been shipped and the dojo threshold has not been met.

Related trainings (3)

Sources (2)

Provenance

Ecosystem: neutral
Added to catalog: 2026-05-27
Last updated: 2026-05-27
Verification status: verified

How the learner advances

Steps (4)

Problem scoping session

Build-deploy-review cycles

Facilitated unblocking

Capstone demo and measurement