hebel. today · 11 November 20266 / 73

Agentic Engineering · AI Agent Management

Shift from prompting to delegating with structured contracts

Effective agent oversight requires moving away from long, repetitive prompts toward formal delegation contracts that define the assignment's objective, necessary evidence, and required human intervention. While many developers focus on the initial launch, research…

Effective agent oversight requires moving away from long, repetitive prompts toward formal delegation contracts that define the assignment's objective, necessary evidence, and required human intervention. While many developers focus on the initial launch, research by Heidelberg and Singapore Management University found that only 0.59% of 36,487 repositories contain the essential engineering for reliable autonomous operation, such as verification and stopping conditions (Super Intelligence School). Teams at scale now treat agent skill updates as code, requiring pull requests that undergo both automated testing and human approval before deployment (Design Systems Collective).

  • Liz Elliott recommends assigning a single, boring job to each agent, using a dedicated context file and a mandatory stop sign for any external actions (Liz on the Web).
  • Delegation requires defining what successful work looks like, as agents may produce technically accurate but functionally useless reports if priorities remain unspecified (Claude).
  • The AI Product Shipping Kit provides 12 agent briefs and a release checklist to help practitioners transition from unverified ideas to pilot-ready autonomous workflows (Ruben Dominguez from The AI Corner).
  • Coding agents function best when managers enforce non-negotiable requirements and maintain human accountability for product decisions and security boundaries (TLDR DEV).

Try today · 30 min

Create a 'Context' file and a 'Stop Sign' protocol for your current agent.

Draft a single document that contains your agent's specific role, its data sources, and the exact triggers that require your manual approval before execution.

You need

  • A text file for agent context
  • List of data connections
  • Defined workflow steps
  • List of restricted actions (send, spend, delete)

Steps

  1. Define the agent's single job in one sentence.
  2. List the specific data sources the agent must read to perform this job.
  3. Write the explicit workflow steps the agent should follow.
  4. Draft the 'Stop Sign' rule: any action involving external communication, financial expenditure, or data deletion requires a human 'Yes' in a specific log file.
  5. Save these as a static context file the agent reads at the start of every session.

One email instead of your newsletter pile.

Forward the newsletters you already get (a Gmail filter does it) and tell Hebel your goals. Every morning it reads them with 20,000 others and sends one short email, with one small thing to do today. Fewer emails, not more. Free during beta.

Start my paper →
LinkedIn X

Sources

  1. Liz on the Web on hebel. →This guide details the four components of a reliable agent: context, connections, workflows, and memory.

  2. Super Intelligence School on hebel. →This analysis of thousands of repositories highlights the missing engineering foundations in most autonomous agent projects.

  3. Design Systems Collective on hebel. →This column explains how teams use pull requests to gate and validate agent-proposed changes to skills.

  4. Claude on hebel. →This masterclass outlines the six-part delegation contract for ensuring agents meet specific professional objectives.

  5. Ruben Dominguez, The AI Corner on hebel. → ✎ Written by a personThis kit offers templates for agent briefs, release checklists, and cost-per-output formulas for pilot deployments.

  6. TLDR DEV on hebel. →This article emphasizes that managing coding agents requires clear context and explicit boundaries rather than endless prompt refinement.

More in Agentic Engineering

Custom agent harnesses versus framework-based orchestration

AI Agent Management yesterday

Design checkpoints replace retrospective review for autonomous agents

Sign in to Hebel first

Likes, your collection and done Hebel need an account. We email you a six digit code; no password.