AI is on both sides of the table

There are two conversations about AI in investment banking and they rarely meet. One is about whether AI will take the analyst job. The other, quieter one is that AI is already sitting on both sides of the recruiting table: screening you on the way in, and reshaping the desk you are trying to reach.

If you are recruiting for a summer analyst seat right now, the second conversation is the one that matters. Here is the honest version of it.

The first interviewer is a screen

Before a human at a bulge bracket talks to you, a recording of you talking to nobody usually gets reviewed first. Goldman Sachs, JPMorgan, Morgan Stanley, Bank of America, Citi, and a long list of middle market firms run one-way video interviews, most of them on HireVue, as the first live filter for summer analyst and full-time analyst applications.

The format is consistent: a question appears on screen, you get about thirty seconds to think, then a short recorded answer, with limited or no retakes. Recruiters review the recordings, and the platforms themselves offer AI-assisted assessment, so whether or not a given firm turns that feature on, your answers are being judged in a structured, at-scale way. Nobody is reading between the lines of a nervous answer, because there is no conversation.

That rewards one specific skill: delivering a tight spoken answer under a clock, alone, with no interviewer energy to feed off. It is a skill you cannot build by reading a guide, for the same reason you cannot learn free throws by watching them. We wrote a full walkthrough of the format in our HireVue guide, including a one-week prep plan.

On the desk, the job already changed

The banks are not experimenting with AI anymore; they are deployed. Goldman Sachs rolled its GS AI Assistant out firm-wide in 2025, to roughly 46,000 employees. JPMorgan’s in-house LLM Suite is available to more than 200,000 staff. Morgan Stanley built assistants that summarize meetings and retrieve research for its advisors. First drafts of pitchbook pages, company profiles, and meeting notes increasingly start life machine-generated.

The common misreading of this is that technical skills matter less now. The opposite is true in interviews. When a model can produce a clean DCF in a minute, a bank has no use for a junior whose only skill is producing a clean DCF slowly. What it needs is someone who can look at the machine’s output and know whether it is wrong: whether the margin assumption is defensible, whether the comps set makes sense, whether the accounting actually links.

So the interview bar on understanding goes up, not down. Walk me through a DCF is no longer testing whether you can do the arithmetic. It is testing whether you understand the mechanics well enough to supervise something that does the arithmetic for you.

What stays the same

Superdays are still humans deciding whether they want to sit next to you at 1am. Technicals are still the entry ticket, not the differentiator. Networking still moves resumes onto interview lists, and a warm referral still beats a cold application. Spoken fluency, the gap between knowing an answer and saying it well under pressure, still decides more offers than any other single factor.

AI did not change what you need to know. It changed how you get filtered on the way in, what the job looks like once you arrive, and, if you use it well, how fast you can get ready.

Prepping with AI: where it works and where it burns you

The case for AI-first prep is simple: the scarcest thing in recruiting is live reps. A friend can grill you for an hour a week. An AI interviewer can grill you every day, transcribe what you said, and grade it, which is exactly the practice the one-way screens and superdays demand. The same logic applies to retention (spaced repetition tuned to what you miss beats rereading) and to markets (a fresh brief generated each morning beats a chatbot with no idea what the tape did today).

Where it burns you, equally concretely:

• Generic chatbots state wrong technical mechanics with total confidence. If a model teaches you an incorrect deferred tax treatment, you will repeat it confidently in an interview. Learn from a curriculum that was audited line by line against the standard IB interview texts, with every worked example recomputed by hand, and use AI for practice against it.
• Copy-pasted AI networking emails read as AI, and analysts delete them on sight. Use AI to research the person and the group; write the three sentences yourself.
• The interview is live and unplugged. Anything you let AI understand for you, instead of helping you understand, is a gap that shows up exactly when it costs the most.

Full disclosure: we build IB Atlas and it is AI-first on purpose, built around the split above. Practice questions can be compared with the product's current reference answers, spoken practice can return an AI study summary, and flashcards can be scheduled around what you miss. Market briefs and networking drafts are also AI-assisted and should be checked and edited before use. The free public beta includes the workspace. And if you would rather assemble it yourself from a chatbot, a flashcard app, and a friend who asks you hard questions, that combination genuinely works too. The tools matter less than the reps.

Common questions

Do banks use AI to reject candidates automatically?

Resume screens have been algorithmic for years, and one-way video platforms offer AI-assisted scoring, but banks generally describe the final review of interviews as human. In practice the distinction matters less than it sounds: your recording is judged in a structured way at scale, so the prep answer is the same either way. Clear, concise, timed spoken answers.

Will AI replace investment banking analysts?

The drafting and formatting share of the job is shrinking, and banks say so openly. The judgment share, knowing whether the output is right and what it implies for the deal, is what the analyst seat is converging on. That is a reason to know your technicals more deeply, not less.

Is it cheating to prep with AI?

No, and the banks are the last institutions that could object, given their own deployments. Using AI to get more graded practice reps is the same as using a coach. The line is letting it think for you in a live setting, or sending its words as your own to a real person.