Test landing page copy by asking 'Do I even know what the product does?'
The host recommends testing your landing page copy by asking 'Do I even know what the product does?' after reading it — if not, rewrite it.
33,353 results · Most relevant
The host recommends testing your landing page copy by asking 'Do I even know what the product does?' after reading it — if not, rewrite it.
The host recommends showing the user the solutions to their problem on the landing page rather than segments you target.
Build agents into your application by putting a ceiling on their permissions, so they operate using an agent token with scoped access instead of taking the user's token or an API key. The three takeaways are: you can let agents into your application with an AuthMD file, connect them to a customer using the user claim flow in AuthMD or the IDJAG identity assertion, and limit agent actions by enforcing permissions and revocation.
When a user invokes an agent, the application's back end should issue agent tokens. Do not just hand an agent the user's token (e.g., Jordan's user token). The reasons are: (1) auditing and attribution, (2) revocation — the user should be able to revoke the agent, and the application should be able to revoke the agent without impacting the user's session and access, and (3) limiting what the agent can do by limiting the scopes in the token. Using scoped agent tokens enables three guarantees: you can audit the agent, revoke it, and limit its permissions.
Commercial context: The creator describes this as their own product.
A key challenge is capturing the vocabulary: you have numerical values, categorical values, and information like zip codes, raising the question of whether to encode each zip code as a token. Push significant effort into understanding the vocabulary you need and the vocabulary you want to scale. If your vocabulary is far too big at training time, it becomes messy to train such a model. Invest effort in deciding which information to encode in a token (value-to-token mapping). A too-huge vocabulary makes training a tabular foundation model messy, so significant effort is needed deciding which values to encode as tokens.
When a call ends awkwardly with just 'thanks for the time,' respond with: 'This is a great use of time. Thanks again for spending all this with us. I've got a couple of ideas. Based on what you shared, there's a couple of different things we can do to help you. I'd love to give you a tour of our technology... and talk about how this can specifically apply to the things you shared with me today. How does that sound?' Don't give a demo or a 98-slide PowerPoint deck — offer a 'tour of our technology' instead. Make every demo bespoke to what the prospect shared: 'I'm not going to give this same canned crap that I've given 19 times today. I'm going to give you something that's bespoke because I actually listen.'
Instead of 'What brought you here today?', open discovery calls with a broad, open-ended question — e.g., 'I could tell you a million things about my company, but I'd love to hear from you first: your team, your challenges, the overall landscape.' This 'cracks their head open': the buyer explains why they showed up in a super professional way rather than narrowly adhering to what the contact form said. Don't fixate on the contact form — acknowledge it, then redirect to broader questions and get to the specific topic later.
After securing the next meeting, ask: 'Who else would benefit from this in your organization? Do you have peers on other teams or in other lines of business — or pending you seeing this and it being successful, anybody else that you think would need to see this before deciding if we want to move forward?' Framing the ask around who would benefit is less salesy and more customer-centric, making prospects more likely to name two or three people. Expect some prospects to refuse ('I'm the ruler of all things. And you have to go through me first') — it's not a silver bullet, there are no silver bullets in sales, but it's a great start.
Don't pepper a prospect with questions for the full 30 minutes — weave in a customer story that mirrors their situation, including specific numbers, then stop and let them react. Structure it as: situation match ('a customer came to us with the same thing'), what you did, how you solved it, and the coolest result — e.g., a final sequence with an 80% open rate and ~7% response rate. After delivering the punchline, pull the rug out and move to the second point — the goal is to make the prospect say 'we want that' repeatedly throughout the call so next steps happen without pushing.
Chris Dreyer advises niching down so specifically that you have a tailored offer: know how many cases you need to create, what content to create, and which marketing channels your niche congregates in. No one says 'I'm in home services' — they say they're a plumber or a roofer; define your niche that specifically.
Gary Vaynerchuk gives a concrete post-production recipe: every live show episode should become a Substack and Beehiiv article and be cut into 19 clips posted on every platform.
Rob Walling's first strategic approach is to take a general tool and apply it to a new vertical, such as scheduling for hair salons or accounting for car dealerships. He notes this has been popular since SaaS emerged in the late 2000s, when low distribution costs made small niches worthwhile.
Alex lays out a giveaway campaign: run a $2,000 private summer camp giveaway as the 'ballistic' version of the campaign, leading with the $5,000 private summer camp full summer experience as the main hook (anchored by starting with the really expensive thing first). Calculate the total value of the package ($500/week x 10 weeks = $5,000) to anchor the prize, and set second place at $3,000 off the $5,000 value, making the ticket $2,000. Make sure more than one person can win second place to stay legal and compliant — say 'more than one person can win' and announce first and second place prizes explicitly instead of saying 'everyone is going to win,' which removes the excitement. Run the giveaway as a seven to ten day campaign with giveaway ads plus a landing page, a countdown timer at the
Start collecting all the ground truth you can evaluate your agent against — from historical data, production, or samples run with your domain experts. If you're in a regulated space, define your hard limits (things that definitely shouldn't happen, e.g., in financial services the agent should never give investment advice) and ingrain them in your evaluation set. Ideally this ensures they never happen; at least you will know about it when it does.
Doneyli De Jesus advises verifying that an LLM judge catches what you intended by manually judging a representative sample yourself and optimizing the judge for coherence with human raters. Without calibration against human-labeled ground truth, the judge will not judge the way a human would and may miss what you wanted to catch.
For customer-facing or regulated-industry use cases, don't risk the agent doing something it shouldn't: build ground truth datasets before production. Curate them with domain experts within the company, include expected outputs (sample replies, expected behavior, keywords), and validate them with customers. Ground truth is also the foundation for auto-improvement.
To ensure LLM-as-a-judge consistency (inter-rater reliability), judge a representative sample yourself, then run your LLM judge and optimize for coherence with what your human raters do. Academically you'd have a few human raters; practically, one human labels some data and you optimize the judge against that ground truth. Check the judge over time to make sure it stays representative. When running on production data, start with a few too many false positives — catch a little more than necessary but don't miss anything — then tune it down over time.
Define quality gates for your agent and verify they are being enforced and scored on every interaction. Scores are automatically calculated whenever a new interaction or trace is captured by the system. Score whether information given to the customer is compliant with banking scope (e.g., a weather question is not compliant), detect failure modes and guardrails blocking answers, and use a score to flag customer intent (e.g., investment advice) so you can easily filter for those cases in traces.
Pamela Fox recommends forcing the model to associate each piece of information in its answer with a source, so it realizes when information wasn't in the sources, and adding explicit prompt instructions like 'if the answer for a bit of info is not in sources, say no source provided' and 'do not make up an answer' if a model answers from its weights in RAG. For RAG, instruct the model to answer based on the sources and cite the sources.
Pamela Fox recommends using DSPy to declare your program's constraints and let it generate the prompt that adheres to those constraints. You check in your spec and let DSPy write the prompt based on evaluations, set up your spec and then set up checks/tests to see if the output satisfies the constraint, and give DSPy the student model and the reflection model, then it runs the optimizer to find the best prompt.
Choose sources → to get a library of your own and a daily brief.