AI ENGINEERING
Free webinar · Online · live · 4 hours

AI agent workflows,
pattern by pattern.

A free live session on how agents are put together. We cover the parts every agent shares, the workflows that need no agent at all, single-agent loops, multi-agent systems, and how to tell whether any of it works.

Message me to join →See the slides
4
hours live
5
modules
21
topics
0
cost
webinar --info
› webinar --info ai-agent-workflowstitle AI Agent Workflowswhen Thursday, 29 October 2026time 10:00 – 14:00 Tehran timewhere Online · liveprice Freehost Pouria Jahandideh› webinar --join✔ message me → added to the Telegram group› ▋

Thu 29 Oct · 10:00–14:00 Tehran time

Reasoning LLMs vs agentsToolsMCP & SkillsRAGContextAgent memoryPlanning & reflectionPrompt chainingRoutingParallelizationReActPlan & executeReflection & self-refiningCodeActSequential agentsMulti-agent parallelizationSupervisorDispatcherEvaluator-optimizerPeer-to-peerEvaluating agents
The thread through all four hours

Who decides what happens next?

Every pattern in this session answers that one question differently. Move the decision from your code to the model and you gain flexibility, and you pay for it in cost, latency and predictability. The skill is knowing when that trade is worth it.

  1. 01

    Workflow

    Your code writes the arrows. Every call does one job you chose in advance.

    inputLLMoutlinegatecheckLLMdraftoutputpassfail: retry
  2. 02

    Single agent

    The model picks the next step from the tools you gave it, until it decides it is done.

    goalthinkacttool callobservetool resultanswerdone
  3. 03

    Multi-agent

    Models pick the next step and which agent takes it.

    usersupervisorsubagentsearchsubagentcodesubagenttests
The agenda

Five modules, 21 topics

We start with the building blocks and climb, one rung of autonomy at a time, to systems of agents. Then we measure them.

01

Agent fundamentals

The parts every agent is made of

  1. Reasoning LLMs vs agentsA model that thinks longer is not an agent. The loop and the tools are.
  2. ToolsName, description, schema, function. The model proposes, your code executes.
  3. MCP & SkillsOne standardises how tools plug in, the other teaches the agent how to work.
  4. RAGChunk, embed, retrieve: adding knowledge at query time without retraining.
  5. ContextThe window is the agent’s whole world on every call. Budget it.
  6. Agent memoryShort-term state, and the long-term kinds: semantic, episodic, procedural.
  7. Planning & reflectionBreak the goal down before acting; check the work before calling it done.
goalthinkacttool callobservetool resultanswerdone
02

Non-agent workflows

Your code owns the control flow

  1. Prompt chainingFixed steps, each output feeding the next, with gates in between.
  2. RoutingClassify once, then send the input down the path built for it.
  3. ParallelizationSectioning and voting: many calls at once, then an aggregator.
inputLLMoutlinegatecheckLLMdraftoutputpassfail: retry
03

Single-agent systems

The model picks the next step

  1. ReActThink, act, observe, repeat: the loop under almost every agent.
  2. Plan & executeWrite the whole plan first, run it, replan when reality disagrees.
  3. Reflection & self-refiningCritique and revise, ideally against a signal the model can’t fake.
  4. CodeActActions as code instead of JSON tool calls, inside a sandbox.
goalplannerstep 1step 2step 3answerfailed: replan
04

Multi-agent systems

Models pick the step, and who takes it

  1. Sequential agentsA pipeline where every stage is a full agent with its own tools.
  2. Multi-agent parallelizationFan out to agents with separate contexts, then merge.
  3. SupervisorOne agent decomposes the task, delegates, and synthesises.
  4. DispatcherTriage the request and hand it to exactly one specialist.
  5. Evaluator-optimizerA worker and a critic in a loop, with a hard stop.
  6. Peer-to-peerNo coordinator. Powerful, rarely needed, hardest to debug.
usersupervisorsubagentsearchsubagentcodesubagenttests
05

Evaluation

Proving any of it works

  1. Evaluating agentsOutcome vs trajectory, pass@k vs pass^k, and a suite built from real failures.
taskoptimizeragentevaluatoragentdoneattemptpasscritique · max N rounds
The slides

Read the deck before you come

The full deck is public: every pattern as a diagram, with when to use it and what to watch for. Arrow keys to move, F for full screen, and print it to PDF if you want it on paper.

Open the slides →
04 · Multi-agent systemsSupervisorusersupervisorsubagentsearchsubagentcodesubagenttests AI Agent Workflows
How to join

Three steps, all of them short

  1. 01

    Message me

    On LinkedIn or at pjsofts@gmail.com. One line is enough.

  2. 02

    Join the Telegram group

    I add you to the webinar group. The live link and reminders go out there.

  3. 03

    Show up Thu 29 Oct

    10:00–14:00 Tehran time, live. Bring questions, especially about agents you are already building.

Pouria Jahandideh, host host
Who is hosting

Pouria Jahandideh

AI Engineer · Engineering Team Lead

I build AI agents professionally and teach the 12-weekAI Engineering mentorship, where students build a coding agent from scratch. This webinar is the map of that territory: the patterns you will meet in every agent framework, and how to choose between them.

RésuméArticlesLinkedIn ↗
Questions

Before you ask

Is it really free?

Yes. No card, no upsell inside the session. If you like it, the 12-week AI Engineering course is linked at the bottom of this page, and that is the only pitch.

How do I get the link?

Message me on LinkedIn or by email. I add you to the Telegram group, and the live link, reminders and slides all go out there.

Do I need to know a framework?

No. Every pattern is shown as boxes, arrows and a few lines of plain code, so it maps onto whatever SDK or framework you use, or onto none.

Will there be a recording or slides?

The slides are already public, linked on this page. Anything else from the session is shared in the Telegram group.

Thursday, 29 October 2026 · 10:00–14:00 Tehran time

Four hours, every pattern,
no cost.

Send me one line and I will add you to the group.

Message me to join →

Want to build all of this yourself? The 12-week AI Engineering course.