@Athena
An agent that generates investment and deal intelligence for women's sports
I tested @Athena's four use cases and most worked very well, but I couldn't verify that one worked. Where gaps exist, they are largely about demonstrating the work rather than doing it.
Precisely scoped. No ambiguity.
An investor in women's sports, media, and entertainment neeeds daily market intelligence, deal tracking, opportunity scoring, and CRM management.
The paper claims existing AI tools are more horizontal and less capable in this sector, but it does not demonstrate it. Run the same week's news, through a general-purpose model versus Athena, to make this comparison concrete.
A standout section. No notes.
Two proprietary scoring frameworks, both fully documented.
An example showing how the scoring math applies to your output pivot by pivot.
Explaining how the thresholds were determined and calibrated.
Very well-designed. Undersold in the paper.
Evaluations are functioning well. I provided a generated one-paper and created an intentionally weak memo for WHOOP and @Athena scored them appropriately, 3.9 and 1.2 out of 5 respectively. Further testing on data-limited outputs, FC Viktoria Berlin at 4.1 and Ariel Investments at 4.4, both scored proportionally lower than polished one-pagers, with deductions correctly attributed to incomplete disclosure rather than flawed logic. The system is calibrated.
Walk through an example with your Judge LLM agent in the paper, specifically the input memo, the output score, the justification, and at least one flagged improvement.
The non-transferability argument needs more depth.
The distinction between what transfers and what doesn't isn't very clear. Specifically, what is non-replicable about these frameworks after they have been written down?
Framing this as a win-win for your next employer is effective. What transfers becomes an operational advantage for your firm, whereas the non-transferable remains your secret sauce.
A time when @Athena's judgement diverges from yours to provide a clear example of the gap of non-transferability. It will also illustrate the power of your judgement and expertise for senior leaders.
Four use cases and a few edge cases
| Digest | Produced a full six-section output from a cold start. Format matched the newsletter reference. Output was tailored to women's sports and covered the correct asset class categories. | Pass |
| Score (Common Cases) | Correct framework applied to the multiple deals I scored. Reasoning explained for each characteristic. Multiplier applied and explained, but missing a consistent data flag. | Pass |
| Score (Edge Cases) | FC Viktoria Berlin (non-US, thin public data), Ariel Investments (imperfect framework fit), and Lusail Sports Arena, where the prompt contained a fabricated figure and an unverified partnership claim. The first two were handled with explicit caveating. The third caught both false claims and corrected them before scoring. | Pass |
| One-pager | Oura and WHOOP both matched the Atlanta Dream reference template in structure. The Investment Thesis and What I'm Paying Attention To sections were distinct and specific to each company. Source discipline was strong on quantitative fields and looser on strategic partnership claims. | Pass |
| Evaluator | Strong memo scored 3.9 out of 5 with specific justified deductions. Deliberately weak memo scored 1.2 out of 5 with a precise accounting of every failure. Data-limited outputs scored proportionally lower. Deductions reflect the proper criteria. | Pass |
| CRM and warm intros | Two network queries returned no data and prompted a file upload request. It is unclear whether the CRM layer was not connected in the version tested, or has not yet been implemented. Luckily, the agent did not fabricate a connection, which is the correct behavior. | Unverified |