ADVANCED · 8 MODULES · 64 LESSONS

Long-Running Agents: The Harness

Build the system that lets an agent work for hours or days across many sessions without losing track, repeating actions, or calling untested work done.

For you if
You've built an agent with tools and need it to work on its own for much longer.
By the end
You'll know how to save checkpoints, resume after a crash, set budgets and approvals, and trace what the agent did each session.
You need
Intermediate Python (3.11 or later) and basic Git. No API key needed.
  1. 01The harness as a system8 LESSONS
    1. Introduction: an agent that works in shifts
    2. Why one session is not enough
    3. Anatomy of a harness
    4. Lab, part 1: state, journal, sandbox, verifier, context and trace
    5. Lab, part 2: the mock and the `--live` mode
    6. Lab, part 3: tools, loop and runner
    7. The naive agent
    8. Project: diagnose the naive run
  2. 02The initializer agent and the feature list8 LESSONS
    1. Introduction: the first day on the site
    2. Two roles, two prompts
    3. From spec to feature list
    4. Acceptance tests define "done"
    5. `init.sh` and the first commit
    6. Guarding the feature list
    7. Running the initializer
    8. Project: a new rule in the spec
  3. 03The progress log and the startup of every session8 LESSONS
    1. Introduction: the shift change
    2. What goes in the progress log
    3. The startup routine
    4. One feature per session
    5. Ending in a clean state
    6. Git as memory and undo
    7. A full run, session by session
    8. Project: a better handoff
  4. 04Durable checkpoints and idempotent resume8 LESSONS
    1. Introduction: the process is going to die
    2. What durable means
    3. The checkpoint after every step
    4. The journal and the dangerous window
    5. Idempotency keys and the outbox
    6. Crash and resume in practice
    7. Durability in agent SDKs
    8. Project: a crash-safe tool
  5. 05Verify before declaring done8 LESSONS
    1. Introduction: the inspector doesn't sign on someone's word
    2. Why agents declare victory
    3. The verifier lives outside the model
    4. Regression checks
    5. Testing like a user
    6. When the agent tries to change the test
    7. Verification gates in agent SDKs
    8. Project: a verifier with one more rule
  6. 06Budgets, bounded autonomy and sandbox8 LESSONS
    1. Introduction: the site has a budget and a fence
    2. Budgets: turns, tokens and dollars
    3. Stopping well: wrap-up and safety net
    4. Approvals and autonomy levels
    5. Stuck detection and escalation
    6. The sandbox: paths and commands
    7. Containers and SDK sandboxes
    8. Project: an autonomy policy for Brindle
  7. 07Context for long sessions8 LESSONS
    1. Introduction: the shift notebook fills up
    2. What fills a long session
    3. Clearing old tool results
    4. Compaction with a harness summary
    5. Compaction in the API and the SDKs
    6. Memory in files, not in the window
    7. Measuring the effect
    8. Project: tune the context policy
  8. 08Observing and debugging long runs8 LESSONS
    1. Introduction: the site camera
    2. The trace as a flight recorder
    3. Reading a session timeline
    4. Following one feature across sessions
    5. Replay: reproduce without a model
    6. Run health signals
    7. From symptom to cause: a debugging playbook
    8. Final project: the ratecard run end to end
Reviews

What students say

These reviews are from enrolled students who completed at least 50% of the course. We moderate reviews only on content grounds (spam, offensive language, personal data), never for being critical or negative.

No approved reviews yet.

Be the first to share your experience!

SEC. 06 · THE STUDIO

Why this studio exists

For decades, building software belonged to technical profiles with years of training, and getting a digital product into the real world took companies months or years. No-code and AI have narrowed that gap, but the knowledge is still held by a few.

Tech Benders exists to open that knowledge up, so business teams and non-technical teams can put their ideas into production with the right technical decisions behind them. AI does what you ask of it. Without a technical base, what it builds may not scale with your company’s ecosystem, and its costs can run away over time.

The team knows the pains of large organizations and the speed at which they have to move. It brings solution architecture, data engineering, product built end to end from back end to front end, and daily work with AI. Software that solves what you need in days, not months, with the cost of AI kept in check.

Founders

The studioRead about Tech Benders →

Write to ushi@techbenders.io

Made with Wislab