Buyside AI Reviews

Former mega fund and credit hedge fund investor writing in-depth reviews of AI tools for finance. Visit the website for the latest write ups, no paywall.

[Buyside AI Reviews @BuysideAIreview]

Notes from today’s Claude for Excel and Claude for PowerPoint webinar:

- Over 10k people on the webinar
- A poll at the beginning of the call showed that half of people who voted have not used Claude or only use Claude chat
- Opus vs. Sonnet vs. Haiku in Excel and PowerPoint? Anthropic team recommends using Opus in Excel and Powerpoint at the moment.
- MCP: if you want to use a certain data connector (ie. CapIQ) in a task, need to be specific in instructions. Claude can help you build custom MCP connectors for any available APIs. But be aware that too many MCP connectors may overwhelm the model.
- Skills: A Skill is a way to teach Claude to remember instructions in a markdown file to be able to run the instructions over and over again for different tasks. Can create this directly in the desktop interface, and it is good to point Claude to templates, brand guidelines, etc. when creating the Skill. Use the Claude “Skill Creator” Skill to help create skills and test them before saving.
- Plug-ins – these are pre-set bundles of MCP connectors and pre-loaded skills.
- Will the result always be the same if I run the same skill? Answer: most of the time yes, but you can get differences in data it finds.

When to use Claude Cowork, Claude Code, Claude in Excel and PPT, Claude Projects?
- Cowork and Claude Code difference is user comfort – if your role does not involve coding, they are functionally very similar in what they can accomplish.
- Claude in Excel or Powerpoint – best to use on several open files that you want to iterate upon (can connect a PPT file to an Excel file within the Excel plugin and edit both). For closed files where you want a finished product without iterating upon it, use Cowork or Code – that's easier for an autonomous first draft. Note: Claude in Word does not exist yet.
- Claude Projects – is limited to a chat construct, good for sharing work instantaneously with team members.

Live demo:
- Very high level AAPL earnings update using CapIQ data using an “Earnings Update Skill”
- “Earnings Updated Skill” is three step: Beat/miss analysis + model roll-forward + research note
- No time to verify outputs, so can't say how well it worked

[Buyside AI Reviews @BuysideAIreview]

@DasWreck Butterfly effect man

[Buyside AI Reviews @BuysideAIreview]

@Clearingfog_ How's the accuracy for you so far?

[Buyside AI Reviews @BuysideAIreview]

Turing test for Excel AI agents

[Buyside AI Reviews @BuysideAIreview]

Have been impressed by the thoroughness of Gemini 3 Pro vs. Opus 4.6 for miscellaneous credit research recently.

I’ve been doing some additional background research on various liability management transactions and Gemini has been great at finding the data I'm looking for without giving up. Below is an example – I wanted to find the cash balance Serta had at the time of the OG uptier transaction (specifically pre-LME cash). Opus 4.6 ran a quick web search and kind of gave up on the task because Serta was a privately held company, while Gemini found a filing from Serta’s 2023 bankruptcy that referenced the cash balance at the time of the LME in 2020.

Don’t see a lot of hype re: Gemini recently but I’ve been finding it pretty useful for research purposes. Curious if others have found the same?

[Buyside AI Reviews @BuysideAIreview]

@the_judge1111 Kind of ironic that many refuse to read AI substack content but think that PMs will want to read AI generated IC memos to save a couple hundred K on analysts...

[Buyside AI Reviews @BuysideAIreview]

@jdmoran https://t.co/49FMsfTptO

[Buyside AI Reviews @BuysideAIreview]

https://t.co/i9RuxhoABK

[Buyside AI Reviews @BuysideAIreview]

This newsletter from @jdmoran does a nice job compiling AI use cases in finance. Great for keeping a pulse on new AI tools and workflows.

The latest edition below mentions my initial test drive of Claude Cowork.

https://t.co/2NtRlBigzc

[Buyside AI Reviews @BuysideAIreview]

@LeylaKuni "Senior" refers to payment priority rather than lien priority. For the same reason unsecured debt is often labeled as "senior unsecured" meaning that payment priority is senior to subordinated debt

[Buyside AI Reviews @BuysideAIreview]

@lolviolence Claude, build me a collateral diligence platform

[Buyside AI Reviews @BuysideAIreview]

@FICMBondTrader Done thx

[Buyside AI Reviews @BuysideAIreview]

@FICMBondTrader haha that's actually gnarly

[Buyside AI Reviews @BuysideAIreview]

@The_Real_Fly The only way to teach them is to make them try to sell a block of bonds

[Buyside AI Reviews @BuysideAIreview]

https://t.co/i9RuxhoABK

[Buyside AI Reviews @BuysideAIreview]

@ManuInvests When Breath Becomes Air. Pretty heavy but puts things in perspective in these crazy times.

[Buyside AI Reviews @BuysideAIreview]

The "why does Bloomberg need to exist" argument comes up every few weeks it seems like and I want to give people a credit investor's perspective as to why Bloomberg has been such a resilient platform over time. There is more to the platform than data aggregation (which itself is a stronger moat than people realize).

People generally love to clown the Bloomberg interface (though the UI works well for credit and I don't mind it given familiarity). But for a large portion of the user base that is focused on fixed income - Bloomberg is an irreplaceable platform for "network-effect" type reasons:

(i) Majority of high yield bonds and all leveraged loans trade over the counter and live bid/ask prices are quoted by dealers many times during a day - these bid/ask levels are only distributed by and to those on Bloomberg terminals.

(ii) Large part of new issue bonds/loans marketing and book building process is done via Bloomberg. Banks communicate with investors during the deal process via Bloomberg.

(iii) Majority of high yield bond and all leveraged loan trades are also executed via Bloomberg (via IB chat + auto trade confirmations) and Bloomberg is routed into the trade settlement systems of dealers and funds (ie. you complete the trade and it flows into fund accounting/settlement departments) - so again, if you are not on Bloomberg, you can't trade with dealers and any new system you use won't be integrated with fund ops.

So in the public credit markets, the terminal is almost a key piece of market infrastructure rather than just a data terminal. To move off of the terminal, you would need to migrate banks + investors simultaneously to a different platform - moving one bank or one buyside shop at a time won't work. For these reasons despite developments in AI I think the Bloomberg terminal will be around for the long time in the fixed income markets at least.

[Buyside AI Reviews @BuysideAIreview]

@ZarathustraCap @getcompoundai is integrated with EDGAR

[Buyside AI Reviews @BuysideAIreview]

Coupon on the EA 2051's is 2.95% which is less than treasury yields today. As a result defeasance (which would require issuer to deposit treasury securities to offset future interest/principal payments on the 2051s) would imply a takeout price below par.

Makewhole price is the greater of discounted payments at T+15 (below par today) and par.

So defeasance is the lowest effective takeout price from the perspective of the Company and gives them negotiating leverage to tender at T+0.

You can take a look at the indenture here: https://t.co/fokmbewmjP

[Buyside AI Reviews @BuysideAIreview]

Interesting thing is that this risk/outcome is a direct function of the high interest rate environment. In prior years (ZIRP) when treasury rates were usually < corp bond coupons, defeasance was not economically feasible (would imply effective takeout price > 101 for issuers) and therefore didn't affect CoC outcomes for creditors. Many credit investors likely hadn't dealt with defeasance risk very often in their careers (I hadn't before) and clearly underestimated the probability of it occurring.

[Julian Klymochko @JulianKlymochko]

In a battle for the ages that pits private equity against hedge funds, it appears that event-driven investors in the bonds of EA are on the losing side.

Those who didn't carefully read the bond indenture and see the defeasance provision will likely receive just 73 instead of the 101 expected under a change of control.