Benjaminsen

I have built similar solutions for multiple clients.

Some basic tips I hope that help you.

1. When the system is up and running don't run agents in a loop, wake them up when something happens. An email arrives, something changed, a timer triggered. There is no reason for 24x7 agents most of the time.

2. You want a system that does not allow conflicts, as your baseline, for projects where many people are working on the same thing. Think git for documents. How is not important, it's important that it's there

3. A basic coding harness like Claude Code is going to do a million times better than anything you can build yourself. It's already optimized for dealing with tasks in a complex environment and gets updated near daily. Don't roll something yourself or buy a special solution from a vendor

4. It's super easy to provide context for a meeting. It's very hard to get people to read them. Spend time making sure pre-reads are short, concise and relevant.

5. Only have one source of truth for core data. An AI agent will happily find an outdated document and assume that's the truth. Make sure it's strongly enforced that important structured data is in an easy to search single database.

Build it incrementally. Pester your CTO about what task is the most important then start there. Spend your time collecting feedback from the people who use the system and build as little as possible. Very easy to add every feature, harder to make sure you only build the features that make the system better.

When that is said, this could easily become a very fun project to work on! Feel free to reach out if you get blocked.

Not a bug as such, but Claude _really_ likes Claudeflare solutions to things that are trivially solved with my own server.

Until I structurally started enforcing defaulting to use my own hardware it was always off on a sidetrack to use some Cloudflare service to solve trivial problems.

Hey thanks for adding in your own feedback!

As you properly noticed by now, I have score but not time. I find adding time decreases the likelihood of people playing again where score is a must.

I would be keen on what you think of my version of the game?

I actually build something to help with the whole pelicans on bike problem! https://modelsalt.com

AI models are incredibly bad at generating high-quality, diverse outputs as a function of how they work.

To the point that I built modelsalt.com in an attempt to create a structured solution to get more diverse outputs without quality loss.

My setup allows bug fixes and improvements where input > output is known quantities to ship anytime. However for anything that has a subjective quality bar I still want to user test it manually to ensure that features I ship to users feels good by my subjective discernment.

I have made 20+ solitaire games in different contexts since 1999 with the latest version being https://dailyplay.com/play/solitaire

Here is cheat sheet to make a game people will actually play

1. The vast majority of players prefer flip one, not flip three.

2. You want the text on your cards so large as you can.

3. People expect that if they click a card, it moves to where it can go.

4. People do not like hard games, give people games that can all be won.

5. You want your pile to sit on the right - at-least for mobile players, reaching across the screen is annoying.

6. People WANT the score. I A/B tested this, people hate when there is no scores.

7. People do not care about sounds.

Things I like from your versions I am totally going to steal.

1. Teach is amazing

2. Showing cards left in the pile

Suggestions

1. The teach arrows are to small, people who play this game have bad eyesight on average.

2. Make teach only show moves that drives the game towards the win state.

3. Add animations, I have AB tested this and the game performs better if you have the basic set

4. Make it load faster, people have industrial attention deficit disorder, you loose 8% of users for each second your game takes to start.

Bugs

1. Reloading the page I can't seem to undo back to the start of the game.

Feature creep is not exactly a new problem, the answer is the same as it has always been -> Only build and keep what you can prove has a positive impact on the product.

Really cool result, in your own assessment how much will you say you as a human provided the creative direction?

Double down on relationships and high quality costumer support. Nobody you want to work with is going to migrate away from your products if they know you well and you are awesome and responsive to work with.

When anyone can build anything it suddenly becomes important who does it. Build trust, build a history of success, build an industry specific brand.

Don't do stuff like squeezing margins by switching to subscriptions...

Super interesting, I run similar experiments in terms of product exploration. For context I run a small one man innovation studio and have all my work happening in the same meta context with agents running in a custom orchestration harness.

Once a week on Friday I throw the following prompt. "Based on what you learned watching me build companies, come up with a unique idea for a startup and build it end to end including getting it live"

Seeing what it creates has been a great way to also help adjust alignment feedback to the overall system.

Here is the first 3 products it build to date:

1. https://whowasalive.com - because it would be good for SEO. (it was right) I gave the feedback that there was little commercialization opportunity.

2. https://everyhexcode.com/ - because it would be good for SEO in a valuable segment, developers and designers I gave the feedback that this was uninspired and built in a crowded space.

3. https://everymcpserver.com - because it would be good for SEO in a valuable segment and gives me an outlet to promote my own MCP servers. I gave the feedback that this is still too crowded a space.

Seeing it keep building generic sites in super crowded spaces I spend some time over this week fixing product and scope alignment. It's currently building today's startup. Will report back when it has figured it out.

Based on what you did here, i am totally going to add "Then build a marketing and self improvement plan you execute on weekly for a month" to that prompt!

Love the visual style!

Taking this to its ultimate final state, I think we will see a world where it's more important who created something than what was created.

This is already true in a corporate setting where a more favored employee will get more credit for the same work as a less favored employee. Or on TikTok where performing a dance will get more likes if the person dancing is pretty.

There will still be an advantage in being able to look at something and say "yes this is good", but everywhere humans are a judge of both the thing created and the person creating it is likely that the person part of the evaluation will be dominant.

So, assuming you work in a corporate setting and as a senior you're already good at the whole "yep this is good" part, spend as much time as you can on your people skills.

Anthropic and OpenAI can't steal your ability to interact with humans... yet?

While we can discuss if AI agents are tools or they should be anthropomorphized, I don't think the lack of credit applies only to AI agents. It's normal that the creative direction setter or even the person who pays for a project is the one to get credited.

Say I don't think Steve Jobs personally built the iPhone or Walt Disney built Disneyland. But that's still how we describe the human that was the core individual in making these things happen.

I would even argue that giving credits to the workers in movies or video games is the exception not the rule.

South East Asia

You do it as a partnership where new employees buy into the company when they get hired.

Compute costs on AI with the same roughly the same capabilities have been halving every ~7 months.

That makes something like this competitive in ~3 years

Zoom supports sharing only a part of the screen if you tab the advanced tab in the screen sharing dialogue.

OpenAI whisper runs on iOS native on device, can't be just costs. https://github.com/openai/whisper/discussions/407

Very similar to prompt fine tuning but orders of magnitude cheaper and works with far less data.

Margins can be turned off in settings.

Pretty sure this will dynamically rewrite the code. No copy pasting needed. We have something very similar at FRVR.ai

Very close to the subsystems we build for FRVR.ai - Although their UX is way cooler than ours, we should get inspired.

I’m preparing. Learning how to work with an AI is the only way to stay competitive. The AIs will become smarter much faster than I will.

Even if that's the case, a sufficiently large hash table is indistinguishable from an intelligence by a lower intelligence.

Really great article. Interestingly I have found that using the function call output significantly improves the coding quality.

However for now, I have not run re-tests for every new version. I guess I know what I will be doing today.

This is an area I have spend a lot of time working on, would love to compare notes.

Whoops just saw the 3D requirement which we do not support yet. For that I would go with Struckd which works on both mobile and pc

We are making a games creator that uses GenAI to make games from human input over at frvr.ai. It’s been a hit with my 10 year old nephew after the initial learning phase.

We show the full source of what you create so you can also use it as a tool to learn coding. (A use case we will support better in the future)

Happy to grand you access if you sign up and submit a developer application at https://beta.frvr.ai/creator/

The isEven package uses the isOdd package…

This happens with protein folding in the body as well, leading to prion diseases such as Creutzfeldt-Jakob disease.

Once the badly folding protein is introduced there is no way back.

https://www.hopkinsmedicine.org/health/conditions-and-diseas...