Claude is getting scarily good

I just dumped a screenshot of a relatively complex RE model from Excel (just screenshot, no Excel file) and Sonnet 4.6 was able to put together the structure and how the inputs link to the model and outputs with relatively little context. It even identified where I was too optimistic in certain tax scenarios.

It gave a pretty detailed review and error-spotting that would've taken an analyst a couple hours just to get up to speed on the model/asset, much less finding errors and directionally wrong assumptions

10 Comments
 

I wouldn't say it's better than a human Analyst (apart from first year DEI hires), much less 2+2 PE associates, but it gets work done directionally right 90-95% of the time way faster

 
Most Helpful

It might be the equivalent going from: 

  • Hand spreadsheets to excel; or
  • Letters to phones; or
  • Physical libraries to internet.

But it doesn't mean we're all gone. 

Imo, deals will get done, the materials to support the deals (models / presos / memos / contracts) will all be of higher quality (more data considered for the output). Maybe deals get done faster. Maybe you need less people to run the analysis. But you do need someone controlling the process and verifying every number (just like you have someone doing that to the work atm). 

It's like a super intelligent intern, with no context other than what you tell it - it'll make assumptions that need to be checked. 

The fastest people I knew at coding were the ones that could code almost as fast as they could think (two keyboards at once). They now code as fast as they can think, without writing code.

The fastest people I know at modelling could model almost as fast as they could think. Now they can.

The fastest writers.. the same.

Basically we're reducing the physical limitations of producing (moving our mouse, typing our keys). The output of our job is the same, but the input steps are compressing.

Those who can strategically plan, think, analyse, to get the output faster with less input (by clearly articulating what they want in the most efficient manner), will be the fastest, rather than how many excel/ppt shortcuts they know. 

We're all going to be responsible for the numbers in our outputs, so now we need to move to systems of efficient planning, instructing and checking of work, with reduced emphasis on efficient 'doing' of work.

 

Reservation at Dorsias

It might be the equivalent going from: 

  • Hand spreadsheets to excel; or
  • Letters to phones; or
  • Physical libraries to internet.

But it doesn't mean we're all gone. 

Imo, deals will get done, the materials to support the deals (models / presos / memos / contracts) will all be of higher quality (more data considered for the output). Maybe deals get done faster. Maybe you need less people to run the analysis. But you do need someone controlling the process and verifying every number (just like you have someone doing that to the work atm). 

It's like a super intelligent intern, with no context other than what you tell it - it'll make assumptions that need to be checked. 

The fastest people I knew at coding were the ones that could code almost as fast as they could think (two keyboards at once). They now code as fast as they can think, without writing code.

The fastest people I know at modelling could model almost as fast as they could think. Now they can.

The fastest writers.. the same.

Basically we're reducing the physical limitations of producing (moving our mouse, typing our keys). The output of our job is the same, but the input steps are compressing.

Those who can strategically plan, think, analyse, to get the output faster with less input (by clearly articulating what they want in the most efficient manner), will be the fastest, rather than how many excel/ppt shortcuts they know. 

We're all going to be responsible for the numbers in our outputs, so now we need to move to systems of efficient planning, instructing and checking of work, with reduced emphasis on efficient 'doing' of work.

I like this and agree with large parts of it. I think this will basically converge to a point where it will be a race for the brain capacity: outputs and products coming back from AI before we (humans) have a chance to review and correct these.

Put another way: historically an MD would provide comments and wait for a day or few hours before receiving them back (especially those that required changes to the analyses). Now, these comments may come back in a fraction of the time.

Say the MD says "add XYZ (obscure) analysis". This will again come back quicker and frankly more such things will be hitting the MDs desk concurrently. I think the bandwidth to review and sign-off will start to matter.

Similarly, people who can be efficient doing the lower level job (numbers) but also scale the salesman role will start making a claim for the MD positions (or senior positions) sooner than they would otherwise have been able to.

From the standpoint of MDs, their key alpha will be their rolodex more than ever. Those that lose this, or let this rust, will be threatened by new challengers sooner than they used to be.

Mid-level people VP / SVP might get expensive "sooner". If an (experienced) associate can handle themself better, then with the efficiency gains of AI they may be able to stake a challenge to the VP / SVP's position sooner.

 

Just wait until banks are charged the actual cost of grinding through Excel nonstop. It's like $20k a month lol. Anthropic doesn't have the VC money to eat that forever.

 

user_5991

Just wait until banks are charged the actual cost of grinding through Excel nonstop. It's like $20k a month lol. Anthropic doesn't have the VC money to eat that forever.

We have unlimited credits on a user basis but must cost them a sht load. I usually burn $600 a month 

 

Et ut modi labore quaerat natus ipsa sequi doloribus. Illum quaerat saepe esse alias iste. Temporibus autem id vel beatae voluptatem magnam. Voluptatem et sit aperiam libero.

Natus autem mollitia explicabo ut nobis esse libero ratione. Est libero facere officia quae rerum minus. Et vel nulla excepturi facilis ipsum adipisci ut. Quia qui qui magni et aut. Corporis et similique vel sequi est. Ut exercitationem qui aut molestias et maxime debitis. Quam accusantium veniam deleniti officiis.

Career Advancement Opportunities

June 2026 Investment Banking

  • Evercore 01 99.4%
  • Moelis & Company 01 98.9%
  • JPMorgan 01 98.3%
  • Guggenheim Partners 01 97.7%
  • Morgan Stanley 07 97.1%

Overall Employee Satisfaction

June 2026 Investment Banking

  • Moelis & Company No 99.4%
  • Morgan Stanley 02 98.8%
  • Evercore 01 98.3%
  • BMO Capital Markets 12 97.7%
  • Banco Santander 01 97.1%

Professional Growth Opportunities

June 2026 Investment Banking

  • Evercore 01 99.4%
  • Moelis & Company 01 98.9%
  • Morgan Stanley 05 98.3%
  • JPMorgan No 97.7%
  • Goldman Sachs 02 97.1%

Total Avg Compensation

June 2026 Investment Banking

  • Vice President (14) $434
  • Associates (44) $258
  • 3rd+ Year Analyst (8) $210
  • 2nd Year Analyst (22) $179
  • Intern/Summer Associate (13) $156
  • 1st Year Analyst (79) $150
  • Intern/Summer Analyst (73) $101
notes
16 IB Interviews Notes

“... there’s no excuse to not take advantage of the resources out there available to you. Best value for your $ are the...”

Leaderboard

success
From 10 rejections to 1 dream investment banking internship

“... I believe it was the single biggest reason why I ended up with an offer...”