Their execution is improving, but their models aren't as good as Anthropic or Open Ai
136 karma · joined May 17, 2026
[...]
Adjustable levels of difficulty: go from very easy - just get the words right - to advanced language requiring college level fluency and grammar.
Build your own scenario: order a coffee in Paris, buy a book in Madrid, board a train in Rome, or even help in a high-stakes operation in a major German metro hospital.
A coach rates your performance and provides suggestions for improvement.
Scenarios are designed to be rapid fire, 3-5 minutes each.
Are you stuck? Ask your AI conversation partner for a hint!
[...]
A legit attempt at making an AI financial advisor for everyday Americans. Most Americans can't afford a human CFP which can cost thousands of dollars a year. We offer a great AI driven alternative for less than $300/yr.
AI obviously is powerful and quick to build, but intellect, intelligence, and creativity will still lead to disproportional outcomes for users. After all, AI still relies on effective prompting to build well.
The real danger, then, is not AI churning out endless slop, it's actually that the slop degrades the perceived quality of actual products and legitimate effort.
Measuring by users is one metric, yes, and a good one. But, quality of output and performance beyond a simple prompt is what will make the difference. Products couldn't only be measured by the number of users, but also by quality and ingenuity.
Lastly, the products of tomorrow will not defy AI, but will build on top of them. AI will build and perform a higher level of compute, yes, but the products will orchestrate AI workflows, rather than deterministic outputs. Guided reasoning, orchestration, and value creation are human qualities, not AI.
We go far beyond a simply budgeting app by providing a solution for your entire financial life. We look at your finances holistically and make judgements that you can rely on.
We've demonstrated multiple avenues of failure with generic AIs and how we solved them efficiently. You can check our blog for more information.
Essentially. We want to take the failure modes out of the user's hands. Arthur, the AI, is an expert at determining what information is needed to answer a question, storing it, and using it for processing. Our suite of tools ensures the AI remains on track and our philosophy of ephemeral arthur is essential to reducing AI context drift and pollution.
At Pendragon we're not here to nag users (we have an anti-Karen clause in our constitution) we're here to help them achieve their goals responsibility and set up a sound plan best for their situations. Whether that's buying a house, a new boat, or saving for college, we help users achieve their goals safely and efficiently.
Its the difference between "You spend too much on dining, you should be putting that money into a HYSA instead" vs "You spent $150 on a dinner this weekend to celebrate landing that new deal. It's slightly over-budget, but you're still well on track with the goals and plans we set up a week ago. No adjustments are needed."
yes, but...although the fundamentals are basically the same that doesn't mean it translates into an actual plan for a user. you're still leaving the hard part up to the user instead of helping them form an actual plan and stick to it.
The paper talked about how an AI informed the usr to build a financial emergency fund. but, when the user lost their job, the AI completely forget it existed. This proves our theory that context management is the key to unlocking the full potential of AI financial advice.
https://pendragon.foxtrotcommunications.net/blog/when-the-da...
You don't necessarily need to provide everything. Arthur (our AI) is smart enough to see exactly which information it needs to answer a given question. but, yes, the more information you provide the easier of a time the AI will have in answering your question. Arthur doesn't guess. if there is crucial information it needs he will ask for it. it doesn't have to be a Plaid hook up, a csv or even a simple user response is a start.
On your third point — you'd need an outcomes dataset — that's true for traditional ML, but it's not how this works. The normative layer is finance itself (life-cycle theory, tax rules, amortization) implemented as deterministic calculators, with the LLM doing explanation and elicitation. The paper under discussion is sort of the proof: the models already give theory-aligned advice with zero outcome training. The gap it found is input quality and statelessness, not a missing training set.
I know you said that you don't want a TOS link, here's something better. This is our constitution. https://pendragon.foxtrotcommunications.net/constitution