Frame the right agent task
Separate work that benefits from an agent loop from workflows better served by deterministic software, automation or a simpler model call.
AI agent development training
Build agents around real work and control
Hands-on training for teams designing AI agents that can use context, knowledge and tools without hiding responsibility, risk or system behaviour.
The outcome
Agent capability that survives beyond the prototype
Choose the harness deliberately
Compare managed and code-first harnesses against the team, platform, control and operational needs of the use case.
Engineer the complete system
Connect instructions, context, knowledge, memory, tools, identity and approvals as one testable delivery architecture.
Evaluate before autonomy grows
Use representative tasks, traces, graders, red teaming and production signals to decide what the agent can do safely and reliably.
Two build levels
Start with one responsibility, then orchestrate only when needed
The programme separates the foundations every agent needs from the coordination patterns that become useful in more complex work.
A / 01
Single agent
Build one clear responsibility well
For teams learning the foundations or automating a bounded task with explicit context, tools, permissions and human intervention.
- Task contract, success criteria and autonomy boundary
- Instructions, context, knowledge and memory choices
- Tools, identity, approvals and failure handling
A / 02
Orchestrated system
Coordinate specialised responsibilities with control
For workflows that need routing, subagents, handoffs or multiple systems while keeping ownership and behaviour observable.
- Routing, delegation, handoffs and shared state
- Orchestration patterns and human checkpoints
- Tracing, evaluation, recovery and operational ownership
Built for
The teams responsible for agent architecture, delivery and operation
Developers
Build agents that use context and tools safely, expose useful traces and remain maintainable after the workshop.
Architects and platform teams
Make sound harness, integration, identity, data, observability and operating-model decisions across agent solutions.
Technical product and engineering leaders
Set realistic autonomy boundaries, quality measures and ownership for agent initiatives before they scale.
Programme
From agent task to evaluated production system
The final agenda is matched to the organisation's preferred platform, application stack, data boundaries and intended level of agent autonomy.
Agent scenarios and autonomy
Start with the work, decision boundary and measurable outcome rather than the technology label.
- Agent, assistant, workflow and automation boundaries
- Task decomposition, success criteria and failure cost
- Autonomy levels, approvals and human intervention
Harness and framework selection
Choose a build path that fits the scenario, team and enterprise environment.
- Managed, low-code and code-first agent harnesses
- Framework evaluation and portability trade-offs
- Model, hosting, state and integration decisions
Context, knowledge and memory
Give the agent the right information without creating an uncontrolled data surface.
- Instructions, context engineering and structured state
- Retrieval, grounding, knowledge sources and citations
- Working memory, durable memory and retention boundaries
Tools, identity and permissions
Connect actions to explicit contracts, credentials and least-privilege controls.
- Tool schemas, Model Context Protocol and API integrations
- Identity, delegated access and secret handling
- Approvals, policy checks, idempotency and recovery
Orchestration and multi-agent patterns
Add specialisation only when it makes the system clearer or more capable.
- Routing, subagents, handoffs and shared state
- Sequential, parallel and event-driven coordination
- Human checkpoints and cross-agent failure handling
Evaluation and production operation
Create evidence for quality, safety, cost and change before expanding agent responsibility.
- Representative datasets, graders and regression tests
- Tracing, observability, red teaming and incident signals
- Latency, cost, release, monitoring and ownership decisions
Programme brief
Designed around the agent decisions your team must own
Scoping identifies the cohort, scenario, harness options, integration constraints and governance requirements before the exercises are finalised.
- Duration
- 2–3 days
- Audience
- Developers, architects, platform teams and technical leaders
- Delivery
- Virtual or on-site
- Focus
- Harness, context, tools, orchestration, evaluation and operation
- Approach
- Hands-on exercises and architecture decisions using realistic agent scenarios
Practice-informed
Taught from the system outward, not the demo inward
Fedorai founder Arafat Tehsin combines solution architecture, applied AI delivery, community leadership and hands-on training. The programme connects current agent capability to the engineering and operating decisions required for durable enterprise use.
About Arafat and Fedorai Questions
What teams ask before booking agent training
Which agent framework or platform does the training use?
The programme teaches durable architecture and evaluation practices, then uses the harness or framework that best fits the cohort. This can include code-first frameworks, managed agent platforms or a comparison across options agreed during scoping.
Does the programme cover Model Context Protocol?
Yes. MCP can be covered as one tool-integration pattern alongside direct APIs and platform connectors. The programme focuses on contracts, permissions, identity, failure handling and operational fit rather than treating one protocol as the whole architecture.
Do participants need prior agent-development experience?
No prior agent framework experience is required, but participants should be comfortable with software development, APIs and basic AI application concepts. The technical depth is adjusted to the cohort.
Can the workshop use our own scenario or system?
Yes, where access, confidentiality and preparation allow. A private engagement can use a representative workflow, architecture or controlled codebase so that the exercises connect to a real delivery decision.
Does the training include governance and security?
Yes. Identity, permissions, data boundaries, approvals, traceability, evaluation, red teaming, monitoring and human intervention are integrated into the engineering curriculum.
Is this tied to one model provider?
No. The programme can go deep on a selected platform, but the core architecture covers model, harness, knowledge, tool, identity and operational decisions that teams should understand across providers.