SEND HELP (more bugs) 🐜
Want to back the stream? The «داعم» membership on YouTube is EGP 60 a month: badges, emoji, and I answer your questions first on stream. Become a supporter
The gist
The stream follows practical work on organizing development through a customizable Command Center and multiple Warp sessions, including a plan to turn its core structure—swim lane, task board, agent hub, and calendar—into a reusable Markdown specification. Much of the review centers on Talkstore and Content Ops: inventorying use cases, inspecting logs and classification, tracing confirmation-gate, action-chip, and reply-layout bugs, and designing merchant actions that respond to conversational context. Early content-writing and product-enrichment attempts were shown, but they did not produce a complete success; UI regressions, missing product images, catalog-processing latency, and unexpected post-execution chips still required verification. Luqman also discusses a proposed on-demand self-testing diagnostic agent with human review, considers plan mode as a possible safeguard before aggressive fixes—while noting that it was not used in the described case—and contemplates bootstrapping regression baselines for later review, though some setup details remained unclear. Authentication, correlating agent outputs, and defining passing results also remain open challenges. Broader discussion includes qualified opinions about coding tools and usage plans, the importance of genuine product value, validating SaaS demand before building, and extending the learning journey beyond coding into shipping, marketing, and selling.
53 moments from the stream
- 0:36:25
Context-Aware Quick Actions in Commerce Assistant Responses
The speaker discusses what he described at the time of the stream as the hardest part of the work: response logic and formatting. He proposes that a commerce assistant detect executable opportunities in a merchant conversation and present context-aware quick-action buttons, such as applying suggested improvements to selected products or showing other products that may need work. He views these buttons as practical calls to action and criticizes the earlier version as generic, disconnected, and lacking contextual intelligence.
- 0:41:28
A Multi-Angle Review of the Content Ops Domain
A request to examine Talkstore’s newly introduced Content Ops domain through code review, debugging, validation, and security checks, producing a report of completed and remaining work. It was explicitly noted that the domain had not yet been personally tested and would be tested manually after the review.
- 0:42:13
A Copy Button to Reduce Repetitive Work
Adding a copy button made transferring text easier and reduced repetitive manual copying, with a mention of automatically opening the next application; the application’s name is unclear in the captions.
- 0:44:15
Fixing the Confirmation Gate and Rethinking Action Chips
The bug was narrowed by separating normal responses from the confirmation-gate flow and examining the available action chips. The speaker argued that chips should be context-specific calls to action rather than generic or random options. The immediate decision was to fix the confirmation gate and verify it later, then refactor the broader chip workflow; no successful fix was claimed.
- 0:52:15
Specialized Agents for Bug Review and Phased Fixes
A debugger agent and a researcher agent were presented as reviewers that could help catch bugs. The work proceeded with a first phase, followed by refreshing the running server and testing a trigger that executes a product-enrichment flow. The server recognized the execution command, but the response appeared slow and the cause remained uncertain, with several possible sources.
- 0:55:27
Catalog Processing Crash When a Product Image Is Missing
A prior failure was described in which catalog processing crashed upon reaching a product with no image, identifying a missing-data edge case.
- 0:56:00
Working with Coding Agents in Antigravity and Cursor
The speaker discussed his experience and opinions of Antigravity and Cursor. He said Antigravity's terminal sandboxing can affect which agent receives a pasted screenshot and argued that these IDEs are designed to encourage use of their built-in agents. He personally preferred Cursor while noting its cost and usage constraints, and said he had only viewed material about a recent update rather than claiming to have tested it himself.
- 1:00:32
Unexpected UI Chips After Execution
While inspecting the result, the speaker questioned why UI chips appeared on the result message after execution, identifying unexpected behavior that needed investigation.
- 1:01:13
Processing-speed concerns for large product catalogs
A performance concern is raised that the workflow is slow and could take a long time for a merchant with around one hundred products.
- 1:01:52
Debugging the merchant action confirmation gate
The required confirmation gate should give merchants clear approve and cancel buttons while retaining the option to type approval. Although an action-preview component reportedly contained those controls, it did not render in the interface, prompting a request to verify the implementation and provide exact review steps.
- 1:05:00
Inventorying and testing Content Ops use cases
The review asks for a complete inventory of merchant-facing Content Ops use cases, followed by instructions to generate and test each one individually. Visible examples include content audits, article inventory and details, collection checks, content-gap recommendations, content generation, and collection descriptions, along with checks of the classifier and the system’s awareness of content-block counts.
- 1:11:09
Testing Blog Actions and Redesigning the Confirmation UI
A workflow combining blog-content writing with a confirmation gate was tested and its interface reviewed. The speaker proposed placing approve and cancel controls inside the agent reply while retaining typed confirmation, and making the action preview clearer and more polished. The blog-writing attempt initially worked, but the flow later crashed, so the result was not a complete success.
- 1:16:49
Enabling an Agent to Inspect Logs and Diagnose Bugs
The speaker discussed training the agent to investigate bugs more independently while retaining human review, using an API chat and inspecting classifiers, logs, text, and conversation logic to detect problems and speed up the work. The next step was to determine whether this approach was feasible and how it could be implemented precisely.
- 1:18:36
Diagnosing a Confirmation-Gate and Reply-Layout Regression
The speaker tried to determine whether the fault was in the blogging action or the confirmation gate, then described a visual regression: the reply first appeared on a white background, but after the approve-and-execute and cancel controls appeared, it changed to black and the layout became distorted. The button arrangement itself was acceptable; the delayed, inconsistent transformation was the main issue.
- 1:21:25
Designing a Self-Testing and Debugging Agent
A discussion of building an on-demand agent that continuously tests and debugs an application, potentially coordinating debugger and validator subagents. The aim is to automate log monitoring and the transfer of results between system components, while noting difficulties capturing and correlating subagent responses programmatically.
- 1:25:15
Expected Text Streaming and Approval-Bar Behavior
A concise UI expectation: text should stream normally, then an approval bar should appear beneath it without transforming the content.
- 1:25:47
Safety Guardrails for Autonomous Bug Fixing
A practical observation that a persistent coding agent may try increasingly aggressive approaches when a direct fix is elusive, potentially fixing one issue while damaging another backend area if it acts without approval. Plan mode is discussed as a possible way to review the approach before execution, with the qualification that it was not used in the described case.
- 1:28:38
Playwright MCP Limits for Login-Protected Testing Flows
Automated testing is constrained because Playwright MCP access requires a login. Manual screenshots are acceptable, but the broader goal is to test UX flows and backend behavior, which requires a reliable communication path between the testing agent and the backend.
- 1:30:28
Defining Success Criteria for a Self-Testing Agent
An incomplete discussion asks what a passing result should look like and how the agent should evaluate a response, suggesting comparison against a known-good baseline. The remaining evaluation criteria are unclear in the supplied captions.
- 1:31:10
Bootstrapping Regression-Test Baselines
The speaker considered structural, semantic, and regression testing, with the agent initially bootstrapping response baselines for later review and issue detection. Some setup details are unclear in the captions.
Got a question about something in this stream?
Ask in the group