O'Reilly Media

LinkedIn

Salesforce

Mike Wheeler

Trained 500K+ students. Teaching Agent Workflow Design & Context Engineering

Cohort 1

Workshop 1

Apply MIT's RLM paradigm without code. Master Agent Sense to decide what agents should handle and Context Engineering to make it work.

Agent Team Architect

Ramp Labs just published their <a target="_blank" rel="noopener noreferrer nofollow" href="https://x.com/RampLabs/status/2042660310851449223?s=20">Latent Briefing: Efficient Memory Sharing for Multi-Agent Systems paper</a>, built on the Recursive Language Model framework this course teaches.MIT's research on Recursive Language Models (RLMs) reframes how AI handles massive context. Instead of cramming information into the model where it degrades, RLMs treat context as an external environment accessed on demand. The result: processing 10M+ tokens while outperforming traditional models and helps avoid Context Rot.The challenge? Building custom massive-context AI infrastructure typically requires dedicated ML engineering teams and significant investment. MIT's RLM research offers a different path: processing 10M+ tokens at roughly $1 per query, outperforming traditional approaches that cost 2 to 3x more and still fail at scale. MIT documents their approach, but turning research into practice has remained out of reach for non-coders, until now.This course teaches you to build your own RLM using Claude Cowork. Cowork's file-based architecture makes RLM principles accessible to non-programmers, achieving ~95% parity with what MIT proposed.<a target="_blank" rel="noopener noreferrer nofollow" class="underline underline underline-offset-2 decoration-1 decoration-current/40 hover:decoration-current focus:decoration-current" href="https://arxiv.org/html/2512.24601v1">Read the MIT paper</a>

MIT cracked context rot with Recursive Language Models. Build a no-code RLM.

Decide what AI agents should automate, augment, or escalate to humans

Understand the RLM paradigm and why context rot breaks AI workflows

Build a working RLM using Claude Cowork with no code

Design context that persists across sessions and survives handoffs

Create stakeholder-ready documentation for agent workflows

Implement internal agents first, then scale to customers

You should be familiar with how work is done today and have visibility into the decisions, handoffs, and outcomes involved.

Experience owning or influencing reaI workflows

This course assumes you are used to making or advising on decisions that affect outcomes, risk, or other people’s work.

Comfort making decisions with accountability

You do not need prior AI or agent experience, but you should want to thoughtfully adapt existing workflows for agent support.

Interest in thoughtfully adapting work for AI agents

Operations and transformation leaders who need to introduce AI agents with accountability but lack technical staff to build from scratch.

Senior ICs and team leads who see AI potential in their workflows but need a framework to decide what to automate versus keep human.

Consultants and advisors guiding AI adoption who need tool-agnostic frameworks and documentation they can defend to skeptical clients.

Salesforce trainer, O'Reilly author

O'Reilly author. 10+ years teaching Salesforce admin, developer, and Agentforce.

Salesforce just shipped Headless 360. Build your first agent in one afternoon.

On April 15, 2026, Salesforce announced Headless 360 at TDX. The platform became agent-native: 60+ new MCP tools, 30+ preconfigured coding skills, a new Developer Edition, and Slackbot as the new control plane for the agentic enterprise.If you're a Salesforce admin, builder, or developer, this changes how your work happens. And if you don't have a mental model for it yet, you're not alone. Most of the ohana is still figuring out what MCP even is.Whether you're becoming a Vibe Admin, Vibe BA, Vibe PM, or Vibe Coder, this workshop meets you where you are. Agentforce Vibes changes how your work gets done. Headless 360 changes what the platform exposes. We'll help you navigate the unknowns together.This workshop gets you hands-on with Salesforce's Systems of: Context Work Agency Engagement You leave with an agent running in your Dev Org, a working understanding of MCP, Agent Script, and Agent Fabric, and a Complete MCP + API + Voice Resources Guide you won't find elsewhere.

Trade the confusion about Headless 360 for a working agent you built yourself.

Build and deploy your first Salesforce agent

Query your Salesforce org through MCP

Modify Salesforce metadata by conversation with Agentforce Vibes

Orchestrate work through Slackbot as the agentic control plane

Map the four systems of the Headless 360 architecture

Leave with a playbook you can keep building on Monday

The recommended path runs entirely in Agentforce Vibes in the browser. Chrome, Safari, Edge, or Firefox all work.

A modern web browser

Short pre-work modules walk through signing up for the free new Developer Edition so minute one is productive, not setup.

10 minutes of pre-work before Tuesday

If you can navigate Setup, you'll move fast. If you're newer, the browser-only path keeps you in step with the group.

Salesforce familiarity helps but isn't required

Salesforce professionals and aspiring vibe admins who are overwhelmed by rapid changes and confused by Headless 360 and Agentforce Vibes

Trailblazers who've heard of MCPs and APIs but haven't wired one up....especially with AI involved.

Anyone leading Salesforce delivery who needs to speak fluently about Headless 360 quickly and know what to use, based on use-cases and reqs.

Claude Cowork: Build Agent Workflows That Pass Audits

AI educator and O’Reilly author focused on accountable agent workflow design

Evaluate the trade-offs: clicks over code, local file access, but no cross-session memory. Know before you build.

Distinguish what Cowork can do, and where it hits limits

Plan triggers based on time, milestones, and remaining capacity. Capture critical context before limits hit.

Design checkpoints that trigger before context runs out

Transform what works into portable .md files you can reuse across Claude Code, Cowork, and other AI tools.

Convert your best prompts into reusable skills files

Design Agent Workflows With Clear Checkpoints

Know which agent actions are safe to automate and which require human review before they create problems you'll own.

Identify AI tasks that need human checkpoints first

Use a simple framework to decide what agents can do alone, when they need approval, and who's accountable when wrong.

Ask 3 questions before giving AI autonomy

Avoid the #1 oversight mistake: giving agents responsibilities you'd never hand a new hire on their first day.

Stop treating AI like a senior employee

Get hands-on with a structured tool that captures intent, escalation paths, and guardrails for any AI workflow.

Use the Agent Design Blueprint tool live

Design AI Workflows You Can Trust

Spot how mixed signals, unclear ownership, and fear of replacement undermine adoption.

Diagnose why AI adoption erodes trust

See how reactive policies create new risks instead of reducing them.

Evaluate why bans and over-delegation both fail

Reframe AI adoption as workflow design with explicit delegation, escalation, and accountability.

Mike Wheeler teaches reliable AI in Agentforce, Claude & beyond

Mike Wheeler

About Mike

Courses

Agent Teams Workflow Design and Context Engineering: Build Agents with Claude

Workshops

Salesforce Headless 360 and Agentforce Vibes Bootcamp: Learn MCPs, APIs & Slack

Lightning Lessons

Claude Cowork: Build Agent Workflows That Pass Audits

Design Agent Workflows With Clear Checkpoints

Design AI Workflows You Can Trust