Finding Your People: What Two Years at McKinsey Taught Me and What AI Is Changing
Summary
- What I came for and found: I learned a lot at McKinsey from leaders who read my work, judged my thinking, and invested in me. But by design, that trust resets with every new study and every new team
- What AI changes: when AI makes polished pages routine, the content that analysts deliver no longer proves understanding. The people who would invest in you have a harder time telling who to invest in, and mentorship becomes harder
- Why I’m leaving: I want to also hone my judgment through results, not just by how people react to my work. Getting that kind of feedback at the AI frontier meant leaving
McKinsey was never part of the plan. In 2023, I was set on becoming a software engineer. McKinsey’s application opened first, so I applied to sharpen my interviewing before engineering recruiting began. Then, unexpectedly, I got the offer.
Over the next few weeks, I messaged a lot of people. I looked for those who had a technical background, chose McKinsey, and later went back to tech. From those dozens of conversations, one in particular stuck with me. He had been promoted faster than his engineering peers, he said, because he was good at the things that engineers didn’t want to do or could not do well.
That was what I was missing. I knew how to build things. But several projects I built stalled or had less impact than I wanted. I never learned how to make work stick through people. I wanted McKinsey to teach me the skills of judgment and influence, from leaders who had already mastered those skills. If it meant double the hours and half the pay, I was willing to take it.
McKinsey at its best
I did gain those skills. I found studies and people that made the work exciting, regardless of however long it took. My favorite clients would often take a lot out of me. I would stay up till 4 AM perfecting the pricing strategy document, but I did it because I felt real pride and ownership over my work. I created the document from start to finish, presenting my work directly to the main VP of the organization. I was accountable for his reaction to the content, good and bad. That work I did for him transformed into something more. The VP would text me directly about the work we were doing at the client. He would pull me into more strategic initiatives, from planning significant long-term capital investment projects to getting invited to meetings with the president of the organization.
Getting there took time. It took a second stint with that client before the team trusted me to lead the work from start to end. The trust came from doing the simple things right during the first stint: building an involved Excel model, creating succinct pages for other people’s sections. I had to show I could learn quickly and deliver consistently before earning something bigger.
Along the way, I stumbled. There was a time my email drafts were too long or when my pages were poorly formatted. Then another time when my Excel model linked the wrong data together, creating a skewed impact estimate. My stumbles grew more significant. The time I had to re-ghost a deck because the storyline wasn’t persuasive. Or rework an analysis because I didn’t anticipate an obvious client objection. Each time, the leadership team gave me direct feedback and showed me how to push my thinking further. Then they trusted me to fix it.
The apprenticeship I found was invaluable. Someone read my work, judged my thinking, and invested in me. I learned to anticipate the team’s reaction before I shared anything. I learned to temper my habit of opening dozens of tabs and disappearing down rabbit holes, and to find what really mattered. I returned to that client because I had earned credibility, and I left with a confidence for handling difficult topics with the most senior clients.
Falling apart and reflecting on the work
After that study, I felt like a baller. I had successfully pushed beyond my comfort zone, created real client trust, and worked on something exciting. I felt like I had figured out my BA toolkit and was well on my way to the next stage.
The next study was the one I was looking for: a client and topic at the frontier of tech. I owned a workstream for one of the business units in this larger company strategy document. It was a real chance to figure out the core driving market trends and strategic opportunities available. Four weeks in, I was rolled off.
I stumbled again. Each iteration on the deliverables had something critical that I missed. I didn’t structure the problem well enough or anticipate the next steps. I kept too many loops open and fell behind the pace. Even though I was asking thoughtful questions in PS and building detailed evidence-backed pages, I was struggling. The team needed someone who could deliver more in a shorter period of time. On the pricing study, the leaders had watched me fix earlier mistakes so a stumble was a reason to coach me. This team had known me for four weeks. The only evidence they had of how I thought were my pages, and the pages showed gaps. Given the pace the client expected, rolling me off was a reasonable call.
When I heard the news, I first felt relief. Each day the client felt like I was digging myself deeper into the hole, where each mistake eroded the already diminishing trust I had with the team. The nights of 4.5 hours of sleep and long distance travel had made an impact on me. The second thing I felt was shame. How had it all fallen apart so quickly? How did I go from the best study experience to then getting rolled off?
The EM and ED were generous in their feedback and time afterwards. I took their feedback to heart, seeking out due diligence (DD) opportunities to get more practice structuring the problem and acting quickly against a set of output. All that practice under pressure further solidified my toolkit. My skills got better, and I took them into every study after. The trust I had built did not come with me.
Comparing these experiences, I came to realize there were a few conditions I needed to perform my best work. The best work had a combination of real ownership, a team invested in me long-term, and depth built from exposure to the problem for a long time. But by design, we transition and switch across so many studies, that these conditions “reset” and have to be earned repeatedly with each new team.
Finding your people and the right studies is hard. Even when by some combination of hard work and luck I found it, that trust between the team and me had to be re-earned. This is the price of broad exposure in a short period of time. It’s how the firm has always worked. But AI started changing how analysts were evaluated.
What AI changes for the analysts
Some think that AI will usher in a golden age for consulting and others think the end of consulting. For what it’s worth, I largely agree with Bob Sternfels. While the nature of the work may change (i.e., outcome based projects, answering more difficult and ambitious client questions), there will still be a high demand for capable consultants who can be hired to help clients answer their most difficult problems in the face of AI uncertainty.
But that is a question I’m less interested in. There is a more immediate question about how AI is changing the experience for the talent at McKinsey. Talent that starts with the analyst class (i.e., BAs, Associates, Specialists). Analysts who choose to work long hours in exchange for client opportunities and to learn from existing firm leaders. Leaders who will teach you how to build your human skills (judgment, agency, vision, influence, etc.) while also giving specific, tailored feedback to your areas of growth.
AI adds strain to this relationship.
As an analyst, gaining mentors and sponsors depends on your ability to demonstrate a strong work ethic and interest in the subject matter. Partners read your deliverables (pages, models), evaluate your thinking, and decide to invest in you. The work you produce signals the quality of thinking and depth of understanding. This is how I found caring leaders and created step-up opportunities rarely available to others. The manual work I did, while imperfect, showed my willingness to learn and put in the work. The gruelling effort was rewarded in the form of real, caring relationships so that I could take on bigger tasks.
With superpowered tools, skill md files, and pre-built firm AI products widely available, deliverables became easier to create. The deliverables became more routinized and commoditized. In the short term, an analyst may be able to dazzle by being a few months ahead of the curve and producing quality content quickly; but the gap between great and average tends to shrink with technology. In a study of 758 BCG consultants, Harvard Business School researchers found that AI improved the work of bottom-half performers by 43%, compared with 17% for the top half. The average analyst became more “productive,” creating more pages and building models faster.
When the work of the analyst was more manual, the analyst created a positive signal for mentorship directly with their work. There was less “polish” in the first iterations, but there was also more room to iterate. The analyst would be given more guidance and direction that would shape their intuition for approaching the work. It was hard to create quality content without developing a clear understanding for the industry and problem first. AI made the “polish” on the day zero answers better, and those initial outputs became the default starting point for expectations rather than more ghosted or scribbled pages.
In recent client studies, instead of showing the old-school approach, managers taught new joiners about the availability of tools and how to use them. There was too much to be done to slow down that kind of old-school development. When I asked a question in PS about how the partners thought about a certain topic, they would point out that ChatGPT could produce a better answer than they could and suggest I spend some time there. Instead of team whiteboarding on the first couple days, I would be furiously prompting alone with an AI tool. Instead of the AP or EM delegating a certain task and showing how to approach it, they now had the tools to do it themselves quickly. Why invest that time in someone else when it could be done much faster now by yourself?
The partners and the managers were right to promote these AI tools, both for the team and using it themselves. The increased presence of “polished” pages among new analysts made it a harder signal to decide who to mentor. But early investment in analysts diminished and the gap between expectation and reality widened. Assuming an EM produced a well storylined document, it was easy to “fill” in the pages with AI generated content and a surface level analysis of the problem. There was less incentive to do the upfront investment in building a real understanding before building pages.
I fell into this trap on one of my DDs. GPT-5.5 had been released, and I was able to create the first few pages I needed on the study within an hour. The content passed my EM’s check, and we presented it as a sample to the partner in PS. The partner saw through the garbled mess I had made and pushed me on simplifying some of the bullets on the spot. I stuttered and meandered. I didn’t understand what I put on the page. In this new industry filled with daunting jargon, I had no inner model for how the pieces connected with one another. I had to redo the work, but fortunately I knew how to do it because I had practice with the “manual” method of building an understanding.
Some of this may be growing pains of adjusting to a new model of doing work, but I also think the incentives make it difficult to both find and provide the mentorship necessary for analysts to develop into distinctive versions of themselves. New analysts were increasingly having a harder time with their first studies because the expectations placed on them did not match what they were being taught. Short term study pressure can be alleviated with AI tools but long term skills suffer. With each new study, the expectations rise for what the analyst is supposed to do with their tenure, but they join under-mentored and under-developed since the tools could substitute much of the “signal” necessary to get by. Simultaneously, the analyst’s work becomes less of a training ground to build those skills because the signal for quality thinking and distinctive BAs becomes more muddied by the high outflow of decent-to-great outputs from AI.
In this environment, the best analysts have to train themselves. Seeking out opportunities to learn AI skills outside their client work and deliberately pushing their own “manual” thinking. But if you join McKinsey to learn from leaders who mentor you, teaching yourself makes the long hours harder to justify.
Chasing different feedback loops
For me, I did find my people. I did find those who care for me deeply. I was finding studies that gave me real step up opportunities to be in the client room and present my work. Why not stick around with the people I found and build upon those opportunities?
Staying would give more opportunities where I’d develop my pattern recognition skills and train my judgment further. But I would be developing those against the reactions that others had for my work. I’d be judging the quality of my work by what my EM’s review of the work was, by the annual review feedback that partners would leave for me, and by the client’s comments in the room. I’d never see the prices go live or see the details of the impact of any strategy document I made. I had moved on before it played out.
I grew worried that judging my work only by how others reacted, and not also by whether the recommendation held up, might open a gap between my understanding of an industry and the conviction I had for what mattered. A gap exacerbated by the time delay of being an advisor, waiting to hear secondhand how impactful the recommendations were. Staying one layer removed reduced my ability to experiment directly, waiting to see patterns form across clients rather than testing more boldly.
This tradeoff is endemic to consulting. The difference now is that learning from proxies for outcomes (e.g., how the client reacted to the deck, how the EM rated your work, semi-annual reviews) might create greater, harder-to-see gaps in real performance. When AI makes polished work cheap, a reaction tells you less about whether you understood something and whether that work is a differentiator for the end impact. The one signal that can’t be faked is whether the recommendation actually worked.
I tried to seek an opportunity where I could have my work graded by results, without leaving. I looked into the secondment offerings, in search of a place where I could work for a 6-12 month period at another organization before returning to the firm. I was surprised to see the restriction against startups. Any organization operating for less than 5 years, earning less than $100M in revenue, or having fewer than 250 employees was defined as a startup. By this definition, Anthropic only recently qualified (founded in January 2021). For almost its entire existence, one of the labs defining the frontier was off-limits. Getting that kind of feedback at the frontier meant leaving.
McKinsey still produces more Fortune 500 CEOs than any other company. But if I take the potential impact of AI disruption seriously, the advantage shifts. Judgment trained by advising many companies is likely to become less valuable than judgment formed inside one, in a faster feedback loop. As one Fortune article notes, tech firms “often elevate leaders who came up through engineering or product ranks and whose credibility comes from building and scaling platforms rather than analyzing them.” Credibility that is earned from making and living with the consequences of a decision. If AI makes the cycle of learning faster, it becomes more important to ship a decision, watch what happens, and own what comes next directly. That is the credibility I want, earned by building and graded by the results.
What’s next and the people
Over the last few months, I’ve been reconnecting with that part of me that loved building new projects. I built my own running app and used it every week. Somewhere between fixing what would break and adding what I wished it had, I remembered how much fun building is. I’m taking that into a new role at a startup in SF, where I’ll find out quickly whether what I build works.
I’m leaving with real gratitude. The people I found here taught me what I came for. More than once, someone put me in front of a senior client before I felt ready, and then helped me get ready. That’s how the timid college graduate I was learned to walk into unfamiliar rooms and contribute meaningfully. I saw the same growth and eagerness for reinvention in the people around me, in the AI workshops I led and in the new analysts I welcomed in their first week.
If you ever told me what my work was missing and then let me try again, thank you. If you were one of the analysts I leaned on during a hard study and became a friend along the way, thank you. You’re the reason this was hard to leave.
Note: client references and details altered to protect confidentiality.
In the spirit of more learning, I plan to write more. If you’d like to read along and see the different mini-projects I’ll try to make, please subscribe! Would love to stay in touch and hear about your experience at work and beyond.