Beyond the prompt: Nine expert tips for building better AI agents

The AI Daily Brief highlights nine advanced techniques for building with an agent-focused environment like OpenAI's Codex, shared by a member of the development team. These tips move beyond basic prompting to focus on creating durable, steerable, and context-rich AI agents. For businesses, this signals a shift from simple AI helpers to sophisticated digital co-workers capable of handling complex, long-running tasks with greater reliability.

The AI Daily Brief shared insights from a blog post by an OpenAI team member, detailing nine practical methods for maximising the effectiveness of AI agents built using environments like Codex.

This isn't about simple prompt engineering. Instead, the focus is on building persistent, reliable systems that can operate over extended periods, collaborate with humans, and handle complex tasks. The environment is presented as a complete workbench for creating and managing AI agents.

Core concepts for advanced agentic AI

The nine tips can be grouped into three key areas: persistence and reliability, human collaboration and control, and enhanced agent capabilities.

These techniques represent a significant step up from the basic, one-shot interactions that have characterised many early AI tools. They provide a framework for creating agents that function more like digital employees, capable of sustained, goal-oriented work.

The shift from single-serving AI commands to persistent, steerable agents has profound implications for how businesses can integrate artificial intelligence into their operations.

From task automation to workflow ownership

This isn't just about making individual tasks faster; it's about handing over ownership of entire multi-step, long-duration workflows. An agent built with these principles could be tasked with "Prepare the weekly sales performance report every Friday" — a process that involves:

• Gathering data from the CRM (Tool access).
• Pulling metrics from Google Analytics (Remote control).
• Storing interim findings (Structured memory).
• Continuing the work over several hours or days if needed (Durable threads).
• Periodically confirming it's still running (Heartbeats).
• Presenting a draft for human review in a shared space (Side panel).

This moves AI from a tactical tool to a strategic asset. For a small business, this level of delegation can free up the operator to focus on growth. For a large enterprise, it enables the creation of robust, scalable automation for complex internal processes, from financial reconciliation to HR onboarding.

Building trust through control and visibility

Enterprises are often hesitant to deploy autonomous systems for critical functions due to the risk of error and the lack of oversight. Concepts like steering and heartbeats directly address this.

Steering allows a manager to guide an agent in real-time, much like they would a junior employee. This human-in-the-loop supervision is a critical intermediate step toward full autonomy, as it builds organisational trust and provides a crucial safety net. The ability to see that an agent is still working on a long task via a 'heartbeat' provides necessary reassurance and system visibility.

A new requirement for AI platforms

For business leaders evaluating AI solutions, these concepts should become part of the vendor checklist. The key question is no longer just "How smart is your model?" It is now "How well can your platform support persistent, supervised, and tool-using agents?"

Organisations that understand and adopt this agentic-native approach will be better positioned to achieve significant productivity gains, redesign core workflows, and outpace competitors still focused on simple AI chatbots and one-off automations.

Adopting an agentic mindset requires a strategic shift in how you view AI and its role in your organisation.

1. Audit for Agentic Processes, Not Just Tasks. Expand your thinking beyond simple automation. Look for complex, multi-day workflows that require research, decision-making, and interaction with multiple systems. Good candidates include lead qualification and nurturing, competitive market analysis, or managing inventory and supply chains.

2. Pilot a Human-in-the-Loop Agent. Start with a manageable project where an AI agent works alongside an employee, not in place of one. Focus on implementing the principles of steering and a collaborative side panel. The goal is to assist and augment the human, using their feedback to refine the agent's performance and build institutional confidence.

3. Prioritise Your Data Structure. The effectiveness of agents with structured memory depends entirely on the quality and accessibility of your data. Begin the work now to organise critical business information into databases, well-managed cloud storage, or a centralised CRM. An agent cannot use what it cannot find or understand. A data-first strategy is an agent-first strategy.

4. Update Your Vendor Evaluation Criteria. When assessing AI vendors or platforms, move beyond model performance benchmarks. Ask specific questions about their support for agentic features. Can their system support long-running processes? How does it handle memory and tool use? What mechanisms exist for human supervision and intervention? The answers will reveal which platforms are truly ready for the next wave of business automation.