User
Write something
Pinned
Welcome to Dash of AI's online community! Introduce yourself to get started!
This is our online community to provide continued AI learning and support, for you to share your goals and progress and to support each other on this journey of super intelligence. Let's get to know each other! 🤖 👏 Comment below and share: 1) Where you are in the world 2) What you would like to learn 3) Something you like to do for fun. 😊 💛
12 Agents You Could Build, In the Order I Would Build Them
Not sure what type of agent to build? 1️⃣ Start here — low risk, high repetition 1. Meeting notes to action items. Paste in messy notes, get a summary, your commitments, and what you are waiting on from others. 2. Customer email drafter. Handles the question you answer forty times a month. You still press send. 3. Product and service description writer. Trained on your own writing samples so the copy sounds like you. 4. Document summarizer. Contracts, reports, long threads. Returns the five points plus anything worth worrying about. 2️⃣ Next — useful but needs your judgment on the output 5. Weekly review partner. You paste in the week, it returns what moved, what stalled, what carries forward. Built for solo operators who have no one asking them these questions. 6. Social post repurposer. One piece of content becomes a short post, a caption, and a newsletter blurb, each in the right register. 7. Proposal and quote drafter. Works from your standard structure and past proposals, leaves pricing in brackets for you. 8. Customer feedback sorter. Paste in reviews or survey responses, get themes rather than a pile of individual comments. 3️⃣ Later — only after the first two groups feel easy 9. Onboarding packet generator. New client comes in, it produces the welcome email, the intake questions, and the kickoff agenda from your template. 10. Competitor watcher. Checks a handful of sites on a schedule and tells you what changed. Needs tool access and real supervision. 11. Invoice follow-up drafter. Writes the past-due email at the right tone for how late it is. Drafts only, never sends. 12. Standard operating procedure writer. You describe a process out loud, it returns a written procedure someone else could follow. The one that makes a business sellable. ‼️ And there are so many more! Choose one of the above or tell me what you want to build and I will give you step by step, how to design your agent. 🏗️ 🛠️ 📠 Your commenting and questions help with content, so don't be shy. (And I am writing these posts and commenting, not a bot- just for your information).
0
0
Agents That Ask You First, and Agents That Do Not
When you build an agent, one of the real decisions you make is whether it checks with you before it acts. Designing with approval (human-in-the-loop): 🙅🏾‍♂️👩🏼‍⚖️ 🛑 🟢 The agent figures out what needs doing, prepares the work, and then stops. It shows you a draft email before sending it, a post before publishing, a change before saving. You click approve, or you edit, or you say no. Term to know: human-in-the-loop, (HITL), meaning a required approval point built into the process. I think of it as being a crossing guard. Designing for autonomy: 🤖 The agent does the same work, but no pause to ask for your approval. It sends, publishes, files, or books or writes code or makes artifacts without asking, and you find out afterward by looking at a log or a summary. It can also run on a schedule while you are asleep. 💤 More on the build and how to decide ... Both are real agents. Autonomous is not a more advanced version of approval-based — it is a different choice, and often the wrong one. How I decide which to build: ask what happens when it gets something wrong. If a mistake means a slightly awkward draft you delete, let it run. If a mistake means a client receives something strange, an invoice goes out wrong, or something gets deleted, keep the approval step. Reversibility is the test. If I can't set something back when something goes wrong, it is not automated. Most useful setups are a mix. The agent researches and drafts autonomously, then stops at the one moment that matters— the send button. YOU get to decide. Personally, I don't allow any emails to be sent unless I approve. I don't always have AI write my drafts, either. Sometimes, writing things out helps me to process a phase of a project to know what I need to do next. This is where you are in the designer seat. You need to understand your processes and when you need or want to step in when building agents. Where in your work would you want the pause, and where would you be glad to skip it? Name a specific task and we can talk through where the approval step belongs. 👇
0
0
What You Are Actually Building When You “Build an Agent”❓🤖
A few of you asked me to cover this, so let’s start with the part most explanations skip. Every agent, no matter how fancy, is four things: 1) Instructions: who it is and how it should behave. (You are my Chief of Staff) 2) Knowledge: the documents or facts it can draw on. 3) Tools: what it can reach — your calendar, your files, the web, other software. 4) Trigger: what makes it run. You asking every time, or a scheduled task. (Inbox brief or Inbox triage) That is the whole anatomy. Everything else is packaging or what we say in AI, a "wrapper". Which of those four would help you most right now? Say which one and we will start there. 🤗
0
0
Starting an AI business - What to build? 🤷🏾‍♂️
I have had a few students and recent graduates ask me lately "how to start an AI business" or "what sort of AI business should they build". I thought I would share my replies here with the community. My advice on building a business in AI is like approaching any business - discover a pain point that needs to be solved, build the solution, test the market to see if it resonates, then devise your go-to-market and scale your growth if it does. The difference in building with AI is that it makes it a lot less expensive to start, and very easy to build an MVP (minimum viable product). But that doesn't necessarily make it a successful business because it is easy. Great tips to figure out what to build: ➡️ Ideate with AI- Tell the model (ChatGPT/Gemini/Claude) that you are interested in starting a business (or expanding your business) but you want to have help in coming up with ideas. Tell it to interview you, one question at a time, to get to know you and to help ideate on concepts. ➡️ What are you missing? Ask the model to help you fill in what you are missing. What are my blindspots and what are the risky assumptions? In what ways will this rapidly changing tech environment affect the business I want to build 3-5 years from now? (Personal note - I ask AI frequently to be a devil's advocate on ideas). ➡️ Find gaps in my understanding by asking me questions. Again, use the model as your strategic thinking partner. Ask it to ask further questions to think through every aspect of your plan or project. Not only will the above be helpful if you are looking to start a business- if you are thinking of expanding your business or changing jobs, the same tips can apply. Questions? Comment below and share with the community.
1-30 of 45
powered by
Dash of AI
skool.com/dash-of-ai-8721
Dash of AI — Online SKOOL community to support you on your AI journey
Build your own community
Bring people together around your passion and get paid.
Powered by