Course + in-person training

From first agent session to shipped work.

Agentic engineering is directing AI coding agents through real work you can review and prove. Start with the basics, pick one small project, and move through clear milestones until you can brief, plan, build, ship, and hand off a working slice.

Follow the milestones → Open the course Two-week FDE ramp Open 30 labs

7 milestonesKnow where you are and what proof comes next. Beginner prepGit, terminal, and language basics for fresh graduates. 32 lessonsFoundations, build, standards, scale, operate. 30 labsSDLC, data, ML, AI apps, agentic workflows.

Milestone path

Use this as the main route. Each step tells you what to read, what to do, and what proof you need before moving on.

0. OrientUnderstand LLMs, agents, tools, context, and model choice.

1. First turnInstall the tool, run a controlled session, learn permissions.

2. Real slicePick one small project with a written done check.

3. ShipPlan, build, review the diff, and capture proof.

4. QualityAdd briefs, evals, context, review, and design standards.

5. ScaleCoordinate parallel work with review gates.

6. OperateHandle discovery, teams, cost, security, and handoff.

ReviewShow brief, diff, proof, risk, and next slice.

Open the step-by-step milestone path →

Course

32 lessons in 6 parts. Use the milestone path if you want the guided route, or expand any part to jump into the source lessons.

Part 1 · Foundations

The conceptual stack you walk in with. Skim if you're AI-fluent; do every exercise if not.

5 lessons

1.1

The AI map

Four layers on one page. Stop arguing about the wrong thing.

1.2

LLMs: just enough

Context window, temperature, open vs closed, small local wins.

1.3

What makes an agent

The loop. If you can't draw it, you don't have one yet.

1.4

Multimodality

When to reach for vision, voice, or video.

1.5

The model zoo

Frontier, open weights, specialty, and the routing decision.

Part 2 · Get set up

From nothing installed to your first agent session.

4 lessons

2.1

What is agentic engineering?

Agents versus autocomplete, and why discipline changes the output.

2.2

Install your tool

Get Claude Code running and take the first turn.

2.3

Your first session

Give a goal, watch the agent work, review the result.

2.4

Staying in control

Permissions, review, and safe operating habits.

Part 3 · Build something real

Pick one small project and take it to proof.

4 lessons

3.1

Pick your first project

Choose the one real thing you will carry through the course.

3.2

Plan before you build

Use plan mode before code starts.

3.3

Build it, step by step

Work in slices you can review.

3.4

Ship it

Verify proof before calling it done.

Part 4 · Core concepts

The standards that make agent work hold up.

5 lessons

4.1

The brief

A task contract, not a vague request.

4.2

Evals

Define done before build.

4.3

Context

Write durable project memory.

4.4

No-slop standard

Make the agent check its work.

4.5

Design discipline

Make UI output intentional.

Part 5 · Scale up

From one agent to a team workflow.

3 lessons

5.1

Orchestration

Parallel work with coherent review.

5.2

Review gates

Quality gates and shipping proof.

5.3

Where to go next

Skills, hooks, MCP, harder projects.

Part 6 · Operating in the real world

The envelope you walk out with. The lessons that turn a working agent engineer into a Forward Deployed Engineer.

11 lessons

6.1

The harness wars

Model is commodity. Harness is lock-in. Build the customer's, not the vendor's.

6.2

Application taxonomy

RAG to chatbot to agentic. Know where the customer sits.

6.3

Coordinating teams

AGENTS.md, HANDOFF.md, decisions/. Multi-agent + multi-human.

6.4

Intel-watch

Trusted voices + local judge + alerting. Stop doom-scrolling.

6.5

The team shape

Solo, small, big. Each is a different engagement.

6.6

Problems in every layer

The bug is never in the layer you're looking at.

6.7

Discovery and scoping

One-page scope. Cheapest insurance you'll buy.

6.8

Communicating to non-engineers

Weekly updates. Monthly exec demos. Lead with the metric.

6.9

Observability and cost

Four-axis observability. Defensible cost budget.

6.10

Security and compliance

Prompt injection, PII, residency, audit, access, vendor risk.

6.11

The handoff playbook

The test of whether you did the rest right.

Live workshop: Agentic Engineering Day

Coming soon · Austin first · 8-12 seats

One day in the editor, with a guide in the room.

Claude Code, Codex, Gemini CLI, Snowflake Coco. One shared lab. Real diffs. Live review. You leave with the brief, eval, review-pass, design, and context artifacts installed as a workflow you can repeat.

In-person cohorts are forming now. Want a seat in the first one? Email hello@aisoft.us and we'll hold you a spot.

MorningBriefing agents and keeping control

AfternoonEvals, review gates, memory, and handoff

AfterYour first-week rollout plan

Run it by practice level and CLI

Learner paths

Fresh grad to experienced engineer.

Fresh graduates keep the project tiny and review every diff. Experienced engineers move faster but still do every gate. Team leads turn the artifacts into standards.

Open the practice run guide →

Instructor guide

Teach the loop, not the tool.

Use the same rhythm in mentoring, workshops, cohorts, and hiring filters: brief, plan, diff, proof, decision, handoff.

Open the teaching guide →

CLI variants

Claude Code, Codex, Gemini, Coco.

The course teaches Claude Code first, then translates the same habits to Codex CLI, Gemini CLI, and Snowflake Coco where the workflow differs.

Open the CLI guide →

BeginnerSmall project, every exercise, human review on early diffs.

ExperiencedReal backlog item, stricter planning, proof before done.

TeamShared standards, review gates, rollout plan.

Live cohortShared lab, room review, personal follow-through.

Practice labs

Pick the pathway closest to where the learner is coming from. Every lab has a local small version first, then a fuller version with warehouse, big data, API, model, deployment, or team workflow pieces.

Data engineering

Pipelines, dbt, big data

Snowflake integration, pipeline explainers, dbt tests, Spark tuning, and streaming monitors.

Analytics

KPIs and dashboard QA

LLM data profiling, KPI narrative analyst, and dashboard QA assistant.

Governance

Quality, PII, contracts

AI data quality agent, PII policy scanner, and data contract checker.

More paths

SDLC, ML, AI apps, agentic

CI explainers, release notes, model evals, RAG evaluation, tool calling, task boards, memory, and handoffs.

Career paths tied to real AISOFT work

The training is portfolio-first. Every path maps a learner background to job targets, proof artifacts, interview stories, and the AISOFT offering the work resembles: AI product development, platform design, enablement, full-stack delivery, local LLMs, and edge AI.

Data roles

Engineer, analyst, governance

Snowflake, dbt, KPI narratives, data quality, PII checks, contracts, and portfolio proof.

Builder roles

Backend, full-stack, AI apps

APIs, CLIs, support routing, log triage, review gates, and usable internal tools.

Team roles

Lead, manager, CTO

Rollout plans, team standards, agentic PR review, runbooks, and measurable adoption.

Local AI

Edge and private models

Local model adapters, offline modes, latency notes, and provider-independent design.

The artifacts: fork these

Starter · Lesson 11

Want a guide while you work through it?

AISOFT works with engineers and teams on real code, one practical session at a time.

Book a free 30-min session →