Harness Engineering: Designing the Operating Layer for Reliable Agents
Aishwarya Naresh Reganti
AI Founder & Advisor to F500s | Ex-AWS
Kiriti Badam
Applied AI @ OpenAI Codex | Ex-Google
Learn harness engineering from experts building harnesses at the frontier!
As frontier models converge, the model is rarely what separates a reliable agent from a broken one.
The harness is the operating layer around the model that manages its context, memory, tools, autonomy, and checks. It is where serious teams are investing right now, and far more than a loop over tool calls.
This course teaches that layer as a set of configurable knobs you learn to set, then customize for your own domain. You will build your own harness and customize a popular open-source one for your use case, learning levers that transfer to any tool.
Current Offers:
1. Live course: 25% off with 25OFF
2. 🎁 Full Stack AI Bundle — $2,950 early bird
Become a full-stack AI builder by year-end.
Get this self-paced course with the bundle using this link.
Problem First AI + the Agent-Native Operator workshop
Advanced self-paced courses: Harness Engineering and Beyond Evals
Hands-on Portfolio Build Labs with selected partners to build your resume + $800–$1,000 in Partner Credits (Deepgram, Arize, NVIDIA, Cursor, Lyzr, Grain, Cartesia, Vapi, Mem0, Composio, Exa).
What you’ll learn
Move from using agent frameworks to engineering and customizing the harness layer that makes agents reliable.
Map every lever a harness turns: loop, context, memory, autonomy, delegation, verification, observability.
Learn the failure each lever prevents: context rot, tool sprawl, compounding sub-agent error.
Build a mental model that transfers to any framework or model.
Dissect layers of context window with a budget: compaction, clearing, and offloading without losing coherence.
How do agents like Codex and Claude carry on really long conversations and tasks.
Add the control, state, and gates a bare plan-act-observe loop lacks (stop conditions, step and token budgets).
Actually implement how skills and plugins via progressive disclosure and just-in-time loading works in a harness.
Delegate for context isolation, specialization, or parallelism. Architecture shouldn't define this, user query should.
Understand how learning models like GPT 5.6, Fable delegate in their harnesses.
Decide what runs automatically and what needs human approval (allowlists, command approval).
Learn how sandboxing, execution environment etc. limit what agent can touch.
Implement objective verification: how features like /goal and /loop work, and when to use them.
Use tracing to find where harnesses break, and track improvement metrics.
Learn directly from Aishwarya & Kiriti
Aishwarya Naresh Reganti
AI Founder & Advisor to F500s | Ex-AWS
Kiriti Badam
Applied AI Lead | AI Advisor | Ex-Google
Who this course is for
Technical product managers directing agent work who need to reason precisely about autonomy, reliability, cost, and control tradeoffs.
Engineers building agents who want to move from wiring up a framework to deliberately engineering the harness underneath.
Founders and builders shipping agent products who want the harness to be a durable advantage rather than an afterthought.
Prerequisites
Basic familiarity with coding or the willingness to learn basic commands.
You will be building your own AI harness with the help of coding agents and Python libraries.
What's included
Live sessions
Learn directly from Aishwarya Naresh Reganti & Kiriti Badam in a real-time, interactive format.
Lifetime access
Go back to course content and recordings whenever you need to.
Community of peers
Our alums are product and engineering leaders from some of the best companies: Amazon, Anthropic, Databricks, Google, Snowflake, Notion, Meta, Microsoft and 130+ other companies
Certificate of completion
Share your new skills with your employer or on LinkedIn.
Direct Slack Access w/ Instructors
All students will be part of our Slack community where they can interact with the instructors
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Course syllabus
9 live sessions • 11 lessons
Week 1
Sep
8
Lecture 1: Foundations of harness engineering
[CORE] Week 1: Anatomy of harness and building one from scratch
[Build] Environment Setup for Assignment 1
[Build] Assignment 1: Build your own harness
[Grow] How companies customize different layers of the agent harness
[Grow] How local sandboxes work in coding agents
Sep
9
[AI Builder Summit] The Anatomy of a Self-Improving Agent w/ Arize AI
Sep
10
[Chai & AI] Community Champions Workshops on Applied AI Research
Sep
12
Week 1: Office Hours
Sep
11
[AI Builder Summit] AI Voice Agents and The State of The Art w/ Cartesia
Sep
11
[AI Builder Sumit] Accelerating Voice Agents from Demo to Production w/ Vapi
Sep
13
[Chai & AI] Session with Aish & Kiriti (Not Recorded)
Week 2
[CORE] Week 2: Orchestrating Agent Work
Sep
15
Lecture 2: Advanced Harness Engineering
Sep
18
Week 2: Office Hours
[Build] Assignment 2: Faster Browser Agents
[Build] Assignment 1 Solutions
[Grow] How Codex and Hermes remember across tasks
Schedule
Live sessions
4 hrs / week
Tue, Sep 8
4:00 PM—5:30 PM (UTC)
Wed, Sep 9
5:00 PM—6:00 PM (UTC)
Thu, Sep 10
12:00 AM—1:00 AM (UTC)
Projects
2-5 hrs / week
Async content
1-3 hrs / week
Frequently asked questions
Maven for Teams
Reimbursement
Get your company to pay
Everything L&D needs: email template, receipts, and certificate of completion.
Get reimbursedPrivate cohort
Run a cohort for your org
A dedicated cohort with a custom schedule and curriculum, tailored to your team.
Book a private cohort