Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
9 contributions to AI Automation Society
Been Doing Lead gen for a while, can use some help
It's been about 2 months since I gave cold outreach a third attempt, 6 months of total failures. Spent around 800$ in total zero clients and only 2 interested leads (in the second attempt). I was targeting Digital Marketing Agenices, firstly only SEO Agenices but got almost zero response. Started with generic offers in first attempt in second one sent emails to about 300 people offers were: 1. I will help you double your onboarding speed 2. I will help you double your brief output per writer The first got 2 interested leads, but unfortunately I burned my infra cuz of bad smartlead warmup. In the third attempt I went with these offers and also tried targeting recruiting cuz almost nothing was working in digital marketing niche. Tried initially with more brief output per writer at the start got all negative replies and concluded that CEOs don't want AI to create brief cuz of not good quality (Sent to about 350 people all negative and almost burnt my infra possibly because they marked me as spam). Also tested faster candidate sourcing although that wasn't well tested Than this oct I tested this offer: I will help you handle 5 more retainers without increasing headcount (Client capacity) sent to 215 leads literally zero replies with 10/10 on seed testing. No interest at all. Can someone please help me here, I mean on my lowest point and will quit if it doesn't work this time.
0 likes • 13d
@Ai Kang agreed, finding one severe and urgent problem first makes the email almost write itself, and it fits Muskan's point about sticking with the onboarding angle that already got replies. @Ahmed Belan checking in after a couple of weeks. Did you get a chance to test the onboarding case study again, or clean the list the way Cynthia suggested? If you paste your current first email here, my offer of line-by-line feedback still stands.
0 likes • 5d
@Ali Abdullah Thank you Ali, agreed. Keeping the same offer for two weeks on a small clean list is what gives a clean read. @Ahmed Belan whenever you run that test, please post the numbers here and I will go through them with you.
Quick favor to ask... Glaido is live on Product Hunt today!
We just launched Glaido on Product Hunt. If you’ve been using it, enjoying it, and have a Product Hunt account, we’d really appreciate a quick upvote. It would mean a lot to me and the team. Support Glaido on Product Hunt We really believe in where we’re taking this. The vision goes beyond speech to text. We want you to be able to use your voice to get things done across your tools. If you haven’t tried Glaido yet, you can get a free month here. Thanks for supporting what we’re building! - Nate
1 like • 6d
@Oscar Zapata Hi Oscar, Nate answered this in the pinned thread at the top. The difference he described is that Glaido is a voice layer, so beyond speech to text it can trigger tools like Claude Code, Codex and MCP servers.
0 likes • 5d
@Alfonso García Galan Good question. I keep it simple. Each project has its own folder with one short CLAUDE.md for the rules and context that do not change, plus one status file that Claude rewrites at the end of every session (what is done, what is next, open decisions). A new session reads those two files first, so nothing depends on chat history. For work that spans projects, I keep one tracker in Google Sheets that Claude reads and updates through MCP. I would like to hear how PAL MCP is working for you on this.
Starting the 7 Day AIS Challenge, set these up first
I read all 49 lessons of the 7 Day AIS Challenge before starting my own builds. What can stop you halfway through a day is a missing account, key or file. Here is everything to have ready. Before Day 1 1. A paid Claude plan (Pro, Max or Teams), VS Code and the Claude Code extension, as listed in Prerequisites and Setup. 2. API keys for Anthropic, Perplexity and Kie.ai, plus a Gmail account to send the Day 1 newsletter. Create the keys the night before so the build does not stop halfway. 3. One trip to Classroom, then Claude Code, to download the files later days ask for. The WAT CLAUDE.md (Day 1), the CLAUDE.md and trigger-ref.md (Day 4), the web design CLAUDE.md (Day 5) and the EA Initialize Prompt (Day 7). Day 2 4. A Firecrawl account and your Firecrawl API key (Dashboard, then API Key). The free plan gives you 500 credits. Day 3 5. The Skill Builder files (SKILL.md and reference.md). If the Skills Library link in the lesson does not open for you, the same files are in Nate's public GitHub repo here https://github.com/nateherkai/a-bunch-of-skills/tree/master/.claude/skills/skill-builder. Credit to David Potter, who has been pointing members to this for weeks. Day 4 6. A Trigger.dev account (the free tier works) and a GitHub account. You will use the same GitHub account again on Day 5 and Day 7. Day 5 7. A Vercel account (free tier, and signing up with GitHub is easiest). The front-end design skill installs with two commands shown in the video, so note them down while you watch. Keep a logo and brand guidelines ready in a brand_assets folder, any brand works for practice. Day 6 8. The Claude Code desktop app, which the lesson lists as required for scheduled tasks, and a project with at least one skill from Day 3. Day 7 9. For the capstone, the Submit Your Capstone lesson asks for three things. A screenshot of your EA folder structure with at least one context file filled in, a 1 to 2 minute video of your assistant running a skill, and the Google Form linked in that lesson. Once verified, you are added to the 7-Day Challenge Graduates module.
0 likes • 5d
@Ali Abdullah Thank you Ali, that is exactly why I put it together. Setting the keys up the night before keeps the build moving.
0 likes • 5d
@Tariq Osmani Great tip, thank you Tariq. One env file from Day 1, with the gitignore entry added before the first commit, matters even more once GitHub comes in on Day 4.
Feeling Lost and Unsure About My Next Direction
Hey everyone, I wanted to share where I’m currently at and hopefully get some advice from people who have been through something similar. I’m relatively new to the AI automation space. When I first started learning, I was really excited about what AI could do. I started building my own project, an AI automation agency, and spent a lot of time experimenting with tools like n8n, Claude, and Claude Code. But somewhere along the way, I started losing direction. Looking back, I think I focused too much on building something technically interesting without properly validating whether it actually solved a painful problem or delivered enough value for someone to pay for it. I've put a lot of time into the project, but I still haven't reached the result I was aiming for. There's also another issue: my Claude Pro subscription has expired, and I haven't renewed it yet. Part of the reason is that I feel my Claude setup has become messy. Over time, the memory, project context, and "CLAUDE.md" files have accumulated too many instructions, outdated information, and different decisions. Sometimes I feel like Claude has lost track of the bigger picture and the quality of its responses has become inconsistent. I've even considered creating a completely new Claude account and starting from zero. But I'm not sure whether that's actually a smart move or just another way of avoiding the real problem. Right now, I'm trying to figure out what I should do next. I'd really appreciate advice from people who have more experience in this field: 1. When you realize your project isn't delivering real value, how do you decide whether to pivot, rebuild, or abandon it? 2. How do you manage Claude's memory, project context, and "CLAUDE.md" files as a project grows? Do you periodically clean everything up or start fresh? 3. Would you recommend creating a new Claude account and rebuilding the environment from scratch, or fixing the existing one? 4. For someone still relatively new to AI automation, what would you prioritize to stop spending so much time building things that nobody might actually need?
0 likes • 13d
Hi Niffeine, this is one of the most self-aware posts I have read here, and the replies above already agree on the core answer. Talk to real buyers before building anything more and keep one short CLAUDE.md instead of starting a new account. Two things inside AIS that can help you pick a direction. 1. Nate's new video pinned at the top, "Every AI Business Ranked from Worst to Best", scores 13 AI business models. Watch it with one question in mind, which of these serves people I can actually reach. 2. The AI Business Navigation course in the classroom section walks through landing your first clients. You also do not need Claude Pro renewed to start the buyer conversations, so you can begin this week at zero cost. Once you have spoken to your first 5 people, post a quick update here. I will check back and help you make sense of what they said.
0 likes • 6d
@Niffeine Maliki checking in on the buyer conversations. Josh's plan above is a great way to start, one type of business, one problem they already lose money on, and five owners. Even two or three conversations are enough to spot a pattern, so please share whatever you hear and I will help you make sense of it.
Help me out: How Are You Structuring Your AI Workflow, Handoffs, and Project Tracking?
Hi everyone, I’m currently rebuilding the foundational work processes for my projects. While everyone’s individual tech stack is different (Claude Code, Codex, Cursor, VS Code, various chat interfaces, etc.), I believe the underlying processes our agents and tools follow should be relatively structured, repeatable, and clearly defined. Right now, I'm trying to rebuild my own process to solve the classic challenges around clear work responsibilities, clean agent handoffs, and knowing where things stand. I’d like to know how you’ve set this up in your own operations. Specifically, how do you handle: 1. Scope & Assignment: How do you define what work needs to be done for a given project, and how do you decide who (or which specific tool/agent) does it? 2. Platform & Agent Responsibilities: How have you cleanly separated what each AI platform or agent does versus what the human-in-the-loop is responsible for? 3. State & Progress Tracking: How do you track where a project is at any given point in time? (I’m currently re-evaluating whether project management tools like Linear are the right fit, or if there's a leaner way). 4. Remaining Work & Milestones: How do you dynamically evaluate what remaining work is needed to hit an objective? 5. Completion Criteria: How do you definitively know when a project or slice is actually finished? 6. Stack Organization: How do you orchestrate and separate your tools to stay aligned and drive projects across the finish line without constant friction? If you’ve built out a structured process that works—whether it's for solo development, client builds, or managing multiple systems—please walk me through how you designed it, what your handoffs look like, and how you keep everything aligned. Would love to hear what’s working for your stack so I can apply those principles and best practices to mine. Appreciate any insights you can share! @Nate Herk @Nate Herk
1 like • 11d
Hi Jonathon, the replies above cover scope, ownership and completion really well, so I will add the one piece I did not see yet, your point 4 on remaining work. What has worked in my own trackers is a short milestone list at the top of the project file, each with its done check written before the work starts, as Tariq said. At the end of every session the agent rewrites only that list, done, blocked with a reason, or next. Remaining work is then simply whatever is not ticked, so nobody has to estimate it from memory. Two things already inside AIS match this closely. 1. Day 6 of the 7 Day AIS Challenge, the self-improving scheduled task. One status file per task that the agent reads before it starts and overwrites when it finishes, so the next run knows exactly what happened last time. 2. Day 7, the 4-phase Executive Assistant framework. It gives a folder layout with projects, decisions and references, plus a decision log so you remember why you chose a direction. Which way are you leaning for the human side, a plain file or keeping Linear?
0 likes • 6d
@Jonathon Tamm checking in. The replies since then all land in the same place. Keep state in one file in the repo, write the done criteria before the work starts, and use Linear only if clients or teammates need to see progress, as Jay and Leon said. If you have tried any of this, please share what worked or what got in the way and I will help you tighten it.
1-9 of 9
Soumik Chatterjee
3
17 points to level up
@soumik-chatterjee-2622
Community and growth operator. Helping members get unstuck with AI automation (Claude).

Active 7h ago
Joined Jul 7, 2026
Kolkata, India
Powered by