Because the other thread got way too long, this is a new monthly thread for introducing yourself and any Q&A you have All the resources are in the classroom, but I have not done a good job structuring it, so feel free to ask if you are looking for anything Cheers Tim

If you haven't seen, Claude recently launched routines. You can now automate respetitive tasks in Claude. How to use 1. You can only use routines in Claude code (but Claude code is as simple to use as cowork) 2. Routines can be local (need your computer on to run), or remote in the cloud via Github 3. FIrst perform a task manually prompting Claude. For example, taking project data and generating a report 4. Then say to Claude "I want you to capture this workflow as a .skill" 5. Create a GIthub repo, link it to Claude with that .skill file 6. Then create a routine that says "Run the skill in this repo" It sounds complex, but its actually very simple. Github is just a cloud file storage platform you link to Claude. I did a quick demo explaining. I have a longer one coming.

The Golden Rule Stop using AI as a search engine. Instead, treat it strictly as a "Data Transformation Tool." To get a high-quality result, you must provide: 1. Background Information: The context of the project or document. 2. Transformation Process: Exactly what you want the AI to do with the data. 3. Specific Output Format: How you want the final result to look (e.g., a table, a list, or a CSV). Key Limitations to Avoid - Context Rot: Do not overwhelm the AI. Giving it a 5,000-page contract instead of a specific 1-page document actually decreases accuracy. - Visual Reasoning: AI is fundamentally bad at math and visual tasks. Never use it for quantity takeoffs or counting items on drawings. - Automation Bias: Don’t be fooled by "pretty" formatting. Just because an estimate looks professional and has 300 lines doesn't mean it's accurate. Always verify. 3 Rules for Assigning Tasks 1. Be Grounded and Specific: Use instructions like "Compare these clauses to my standard terms" instead of abstract ones like "Review this contract." 2. Ensure Easy Verification: Only delegate tasks where a human can quickly check the output for errors. 3. Low "Blast Radius": Only use AI for tasks where a total failure won't severely damage or derail the project. Best Use Cases for Construction - Data Transformation: Extracting messy data from subcontractor quotes into a clean bid comparison table. - High-Volume Pattern Matching: Handling tedious tasks, like checking if a finished estimate aligns with every drawing requirement. - Error Checking: Using AI as a "second set of eyes" to spot missing activities in a schedule or errors in your completed work. - Admin Automations: Knocking out "should do but don't" tasks like meeting minutes or categorizing daily emails. Advanced Tips for Success - Deconstruct Complex Tasks: Break large workflows into smaller, individual steps to prevent errors from compounding. - Prompt for Clarity: Tell the AI: "Ask me clarifying questions before you begin the task." - Demand "Brutal Honesty": When asking for a review, tell the AI to be "hypercritical" or "brutally honest." This prevents the AI from simply flattering you with a positive but useless response.

Job seekers also get confused about: including an objective (don't use a summary instead), how far back to go (last 10–15 years only), hobbies (skip unless directly relevant), listing every skill (no, only those matching the job), freelance work (list it as a job with "(Contract)" next to title), references (never include or say "available upon request"), using the same action verbs (vary them try "built," "cut," "launched"), mismatched job titles (use real title + clarifying line like "equivalent to manager"), salary history (never include), and whether to resubmit after a typo (yes, immediately with apology and corrected file).

I'm currently working in a data center, with a target completion date of 1 year, a building 100 meters high. Could you share your experience and knowledge with me on how to manage this project?