You're using AI agents wrong

Balancing Engineering and Business

Struggles with Time Management

• The speaker confesses to being less engaged in engineering due to increased responsibilities in managing a company and YouTube channel.

• Although the speaker's coding skills haven't deteriorated, they struggle to find time for coding amidst other priorities like emails and filming.

• Despite enjoying their current job and life, the speaker misses the satisfaction of coding and problem-solving.

Contribution Metrics

• The speaker shares metrics from their contributions to T3 Code, noting a significant increase in pull requests (PRs), peaking at 52 PRs in one week.

• They attribute this productivity spike to finding effective workflows that allow them to code while managing multiple companies.

Revealing Workflows

Sharing Productivity Techniques

• The speaker plans to demonstrate their workflow live while making real changes during the video, expressing both excitement and nervousness about revealing their process.

Importance of AI Tools

• Emphasizing the role of good AI tools, the speaker notes that many recent changes in T3 Code stem from experiences aimed at improving shipping efficiency.

Browser-Based Agents

Limitations of Current Models

• The speaker discusses how existing models lack access to over 80% of web content without proper browsing capabilities.

Introducing Browser Base

• They introduce Browser Base as a solution that allows agents full access to web resources beyond standard APIs, enhancing functionality for tasks like data fetching or testing applications.

Effective Coding Practices

Workflow Optimization Steps

• The speaker outlines their coding process: starting new threads quickly using hotkeys and ensuring work is not bottlenecked by local machines.

Utilizing Remote Machines

• They describe using an overspec Linux machine for running code changes efficiently without hindering other tasks on personal devices.

Enhancements in T3 Code

Improving Development Experience

• The speaker mentions improvements made within T3 Code itself that facilitate easier multi-threading and remote builds for developers.

Addressing User Issues

Identifying Bugs

• Acknowledging user-reported issues, the speaker aims for simple fixes that enhance user experience without causing confusion.

Setting Clear Expectations

• They emphasize clear communication with team members about expectations regarding bug fixes, prioritizing simplicity over complexity.

Managing Multiple Threads

Avoiding Distractions

• The importance of ignoring threads until completion is highlighted; constant monitoring can lead to decreased productivity.

Motivation Maintenance

• Strategies are discussed for maintaining motivation by minimizing potential frustrations associated with failed tests or broken features.

Auditing Pull Requests

Prioritization Strategy

• The need for thorough audits before merging PR changes is stressed; concerns include API call frequency and reliability of fetched data.

Exploring Other Implementations

The speaker encourages comparing implementations with similar features from other open-source projects for better insights into best practices.

Leveraging Linux Over MacOS

-The advantages of using Linux over MacOS are discussed; Linux provides better performance when handling multiple parallel tasks compared to MacOS limitations.

Audit of Provider Usage Dashboard

Insights on Subscription Limit Meters

• The audit for the provider usage dashboard is complete, revealing significant architectural issues with PR8327 related to subscription limit meters.

• A critical flaw was identified where the API fails to report a 429 error due to an unlimited window fabricated by the system.

• The model highlights that it pulls cloud endpoints faster than community standards, ignoring local data sources that could prevent unnecessary network calls.

Issues Found in Cloud Code

• The model successfully identifies risks associated with 429 errors and recognizes user-reported issues on Cloud Code's closed-source repository.

• Users have experienced delays in checking their usage limits, sometimes exceeding 30 minutes, indicating a serious risk in the current setup.

• There are three major concerns regarding the frequency of checks set at every 60 seconds, which is deemed too aggressive compared to the community standard of 180 seconds.

Drafting Comments for Contributors

Providing Constructive Feedback

• A request is made to draft a useful comment for contributors based on insights gathered from the audit findings.

• The aim is to simplify complex problems and provide clarity on broken skills within Claude's interface.

Understanding Skill Application

• Clarification is sought regarding SLK skills being applied only at the front, raising concerns about user experience and potential confusion.

• Suggestions are made to indicate restrictions clearly when using slash commands in Claude code.

User Experience Improvements

Enhancing Clarity in Command Execution

• Questions arise about improving UX so that users aren't confused by behaviors specific to cloud code.

• A comment drafted by Claude Fable reflects understanding and addresses key issues before merging changes into PR.

Addressing Rate Limit Concerns

Auditing Assumptions Against History

• An assumption about rate limits prompts a request for auditing against personal history due to differing setups affecting data access.

SDK Response Limitations

• It’s noted that SDK responses may not include certain models anymore, emphasizing the need for accurate data representation without unnecessary API calls.

Automating Comment Processes

Streamlining Contributor Communication

• Trust is placed in automated systems to apply necessary changes and leave comments efficiently without further manual intervention.

Managing Open Threads Effectively

Workflow Optimization Techniques

• Strategies are discussed for managing threads effectively while ensuring important updates remain visible without cluttering workspace.

Implementing AI Review Bots

Enhancing Code Review Efficiency

• Introduction of an AI code review bot aims to streamline feedback processes on pull requests (PR), reducing manual effort significantly.

Babysitting Skill Development

• A "babysit" skill has been created allowing agents to autonomously monitor PR comments and implement necessary changes based on automated reviews.

Identifying Friction Points

Improving Project Workflows

• Emphasis is placed on identifying friction points within workflows as opportunities for improvement, enhancing overall project efficiency.

Strategies for Managing Code Changes

Effective Release Management

• Stable releases are safeguarded against regressions through extensive nightly testing, ensuring issues are identified early.

• Alternative strategies include slow rollouts to a small percentage of users, allowing for quick rollback if errors arise.

• A formal QA team can also be employed to conduct thorough pre-release testing.

Derisking the Merge Process

• Key takeaway: Implement mechanisms that prevent changes from affecting many users and facilitate easy detection and resolution of issues.

• Developers should only review pull requests (PRs) when there is high confidence in their readiness, minimizing anxiety about potential breaks upon merging.

Building Confidence in Merging

• Reduce friction before entering PR reviews to enhance confidence that changes will work as intended.

• Address both the perceived uselessness of PR processes and the risks associated with merging to create a safer environment for developers.

Importance of Safety Nets

• The T3 codebase incorporates numerous tests, emphasizing safety nets over mere guardrails; this approach allows for rapid merging without fear of widespread impact from failures.

• Systems built around verifying changes enable faster shipping while maintaining confidence, benefiting both part-time and full-time developers.

Continuous Improvement and Feedback

• The discussion serves as a reflection on personal productivity improvements rather than solely promoting T3 code; it highlights problem-solving through workflow enhancements.

• The speaker expresses a desire to revisit these insights in six months, acknowledging the evolving nature of development practices and encouraging community feedback on workflow optimizations.