David Cummings on Startups

Tag: chatgpt

  • Use Advanced Prompts for More AI Value

    Last week I was at dinner with a group of executives, and of course, the topic of ChatGPT and AI came up. We went around and shared some of our favorite use cases for the technology, as well as ways we are experimenting with AI in our respective companies. Then I asked if anyone used existing prompts or more advanced context engineering. Not a single person did.

    On the one hand, it is so powerful that you can ask questions, give it scenarios, upload documents, have it search the web, and with very little context, it quickly provides valuable answers and insights. So if you can do that with limited instructions, imagine what it can do with much longer, more detailed instructions.

    That’s where prompt engineering templates come in. These are simply text—nothing more, nothing less—but by using these more comprehensive directions, you can get even more value from the AI.

    Let’s look at a shared prompt example from Josh Kopelman for comparing board meeting decks. The prompt asks for two PDFs and does the following, with the output as a PDF document formatted and ready to go:

    1 Snag the Numbers

    Search every slide for hard metrics (ARR, revenue, burn, hires, NPS, runway, churn…).

    Always show Prev (Actual) ➜ Curr (Target) ➜ Curr (Actual) ➜ variance.

    2 What They Said vs. What They Did

    List each commitment from the earlier deck (note slide). Under each, add status & note from the later deck,

    3 Goal-Drift Radar

    Call out any KPI whose name, definition, or cadence changed. List metrics that existed in the earlier deck but vanished in the later one.

    4 Overall Verdict

    • One-liner verdict with a traffic-light word (RED / YELLOW / GREEN).

    • Follow with a breezy 4-to-6-bullet executive summary (≤ 150 words).

    5 Rays of Sunshine & Storm Clouds

    • Top 3 positives (good surprises, momentum).

    • Top 3 concerns (lags, risks).

    Cite slide numbers in parentheses.

    6 Eight Smart Questions for the Team

    Short, pointed follow-ups to close any info gaps.

    Here’s the shared prompt board deck comparison prompt with the text for ChatGPT.

    The overall idea is producing an experienced analyst report that might take a few hours and doing it nearly instantly with AI.

    For entrepreneurs, the recommendation is to search for shared prompts online in the context of the work being done. If you’re analyzing term sheets, go online and find example prompts that do a much more detailed analysis than just a basic prompt. If you’re analyzing a partnership agreement, use the shared prompt that will give you more valuable insights. The AI results are great, and with a more advanced prompt, they’ll be even better.

  • Three Personal Use Cases for ChatGPT and Grok

    Over the last couple years, I’ve been working on incorporating AI tools into my personal workflow. One of the fun questions I enjoy asking other entrepreneurs is how they’ve integrated AI into their own lives and businesses. This has provided me with a variety of ideas and use cases to experiment with and explore. Today, I want to share the three most recent ways I’ve used AI.

    1. Preparing for a Board Meeting

    Last week, I attended a board meeting with 30 other board members. Prior to the meeting, we received a PDF containing the board agenda, a list of attendees, and governance items. I uploaded this PDF to ChatGPT and Grok, asking each tool to extract all the names and companies listed, then search the web for two to three news items or recent events about each person or company. Within a couple of minutes, they generated a bullet-point list of all the attendees, their companies, and relevant recent news. As a result, I entered the board meeting much better prepared, with a variety of topics to discuss during our dinner session.

    2. Analyzing a Potential Geographic Expansion

    One of our companies is researching the next city to expand into, based on our demographic profile and the most common types of users. I fed the details into ChatGPT and Grok, explaining the situation: “We’re considering City X. Our typical users are these types of people, in these types of industries, with these characteristics. Analyze the most prominent firms in those industries in that city, identify the neighborhoods where they’re based, and rank the top 10 neighborhoods based on these parameters.” A few minutes later, I received detailed reports with citations and a strong recommendation for the best geographic location.

    3. Understanding Neighborhood Dynamics in South Downtown Atlanta 

    We wanted to analyze the neighborhood dynamics around residential units and retail use in South Downtown Atlanta. I explained the context to the AI—our neighborhood redevelopment goals, our desire to create a large, innovation-focused district for entrepreneurs and startups, and how residential units might influence the area. I asked: “How would residential units for people living in the neighborhood impact demand for restaurants and retailers?” The AI instantly returned a retail demand equation, calculated based on the number of proposed residential units. It included a range of retail square footage demand and an estimated annual spend per resident at local establishments. In moments, I had a framework and example data to work with.

    These three recent examples really highlight the power of having a thoughtful, personal assistant that can perform deep dives almost instantly. The possibilities for use cases seem endless. I’m committed to further incorporating AI into my daily life and actively seeking out more applications. For me, these examples underscore the value of having instant research and analysis always on demand.