This is a submission for the Runner H "AI Agent Prompting" Challenge
What I Built
As a developer and technical writer, I wanted to go beyond casual experimentation and evaluate Runner H using a real-world workflow. My goal was to assess how well it could:
- 🔍 Find legitimate remote technical writing jobs on platforms like GitHub, DigitalOcean, etc.
- 📝 Draft personalized application emails using my name.
- 🧾 Summarize each company’s tone, target audience, and writing style in a Google Doc for review or future preparation.
- 🛠️ Scan GitHub repositories to identify beginner-friendly e-commerce product page projects I could practice coding with (HTML, CSS, JavaScript).
- 📂 Save project source code to a Google Doc for easy access and provide a breakdown of the interrelationship between HTML, CSS, and JavaScript.
- 📱 Execute all of the above from a mobile device.
I conducted this experiment on both desktop and mobile to compare performance.
Demo
I thought i could embed video directly here. But that didn't work pardon me.
How I Used Runner H
🖥️ Desktop vs. 📱 Mobile Testing
While I didn’t fully analyze the desktop version, I successfully created and executed a prompt before switching to mobile. Since I primarily access the internet through my phone, I was especially curious to evaluate Runner H’s mobile capabilities.
🚀 First Prompt Attempt
Here’s the initial prompt I used:
Prompt
“Search for legitimate, remote technical writing opportunities at developer-focused software companies — including U.S.-based and global platforms like DigitalOcean, GitHub, Vercel, and similar. For 3 promising roles, generate tailored application emails using my name 'Goodluck Chukwuemeka'. Save the drafts in Google Docs. Email me the link for review. Furthermore, summarize each company’s product, writing tone, and target developer audience in a Google Doc so I can prepare if they reply.”
✅ Results:
- Runner H clearly broke down and marked each prompt component as “complete.”
- Clicking “Open File” led to a PDF summary containing a link to the documents — but the link was unresponsive.
- I had read that the app records a video of each prompt session. However, no video was generated. (Is this feature coming soon?)
- Since the PDF link didn’t work, I checked manually via Google Docs and Gmail — the application drafts and company summaries were indeed created.
🔎 Note: All proposals were stored in one document, prompting concerns about whether it might accidentally send multiple proposals to the same company. To investigate further, I created a second, more specific prompt.
✍️ Second Prompt – Enhanced Specificity
This prompt tested if Runner H could generate separate, editable documents (not PDFs) and include functional links.
Prompt
“Search for 10 legitimate, remote technical writing opportunities from developer-focused software companies. Generate tailored application emails using the name ‘Goodluck Chukwuemeka’. Save the drafts in separate Google Docs — one for each company. Additionally, summarize each company’s product, writing tone, and target audience in a single overview document. Finally, create an editable summary (Word format, not PDF) of all actions taken and provide clickable links to each document, all accessible within the Runner H platform.”
✅ Results:
- The system delayed response and gave no feedback on individual task completions.
- It successfully crawled several websites before abruptly stopping. I had to terminate the session manually.
After reviewing the outcome:
- 🟡 Only 7 proposals were generated (instead of 10).
- 📄 One Google Doc was created containing an overview of all companies.
- 📎 The summary PDF again included unclickable links, making it unusable for practical review.
🔍 Insight:
To get usable results, prompts must be detailed and explicitly structured. However, Runner H currently defaults to locked PDF formats, which are not ideal for technical writing or workflow editing.
🧪 Third Prompt – Programming-Focused Test
I then tested Runner H’s technical capabilities with this development-related prompt:
Prompt
“Search for beginner projects in GitHub that I can utilize for developing e-commerce product pages. Provide the source code, explain how the HTML, CSS, and JavaScript components interrelate, describe the project’s purpose, and compile everything in a Google Doc with a shareable link.”
✅ Results:
- It successfully found a GitHub project and provided the repository link.
- It offered a general explanation of the HTML, CSS, and JavaScript relationship — but did not extract, explain, or provide code I could copy and use.
- The Google Doc it created was empty, despite the provided link.
Use Case & Impact
📱 Mobile Experience Breakdown
1. Missing Prompt History
Runner H did not record a video or log of the mobile prompt session, meaning I had no way to trace the session or replay it — a key limitation for repeatable workflows.
2. Uneditable Output
All generated documents were locked PDFs with non-clickable links. For technical writers, this is a major issue — we need editable and interactive formats.
3. Broken or Empty Links
In the GitHub task, the app claimed the analysis would be in a Google Doc, but the link was either broken or the file didn’t exist.
4. Superficial GitHub Analysis
While Runner H located a relevant repository, it didn’t extract or explain the code in a usable way — missing the core of the prompt’s intention.
5. Integration Wasn’t the Problem
All of my Google services (Docs, Gmail) were fully authorized. The issue stemmed from Runner H itself.
✅ What Runner H Did Well
- Accurately understood the general intent of complex prompts.
- Successfully identified and linked real GitHub projects.
- Took initiative in automating workflows.
- Indicated early promise for AI-driven task chaining.
🛠️ Recommendations for the Runner H Team
- Enable Prompt Video Logs on Mobile: Users should be able to trace or replay prompt sessions.
- Provide Editable Output Formats: Avoid locked PDFs; default to Google Docs or editable Word files.
- Fix Link Embedding and Sharing: Ensure all shared files are properly accessible and clickable.
- Add a Review Step Before Email Sending: Allow users to preview and edit generated emails.
- Enhance Prompt Execution for Developers: Provide real, usable code and contextual explanations — not vague overviews.
These small improvements could significantly enhance the mobile experience and build long-term user trust.
💡 Final Thoughts
I walked away from this experiment with valuable insight:
- I tested Runner H under real-world conditions, applying both developer and technical writing perspectives.
- I didn’t just try to make it work — I tried to break it, to understand its limits.
- In doing so, I provided feedback that can directly inform platform improvements.
This wasn't about finding flaws, but about pushing boundaries and contributing to meaningful growth.
Thanks for reading.
Top comments (0)