Breaking Business News | Breaking business news AM | Breaking Business News PM | Business News Select | Link from bio | Quotes | SMPostStory

Quote: John Waldron – Goldman Sachs

“Ultimately, I don’t know what [AI] means in terms of our headcount and how many people are doing which jobs. I’m sure there’ll be some new jobs that get developed, but one thing I’m pretty convinced of is that our really capable, smart people are going get more capable… Hopefully that means our client relationships become stickier, the value proposition becomes stronger.” – John Waldron – Goldman Sachs

Large banks are treating artificial intelligence less as a discrete project and more as a pervasive change in how work is organised, priced and delivered to clients ¹. In this shift, the most consequential decisions are not about which model to deploy, but about how to reconfigure the relationship between scarce human expertise and increasingly capable machine systems. John Waldron’s comments sit precisely in this tension: the technology is advancing faster than any credible workforce forecast, yet the competitive edge of a firm like Goldman Sachs still rests on human judgement, relationships and trust, not just on compute.

In practice, this means senior leadership must hold two apparently conflicting ideas at the same time. On the one hand, there is genuine uncertainty about how many roles will be automated, reshaped or created as AI diffuses through trading, investment banking, risk, operations and wealth management. On the other hand, there is a very specific conviction: the people who are already top performers will gain disproportionate leverage from these tools, amplifying their impact with clients. The quote does not duck this contradiction; it treats workforce numbers as a contingent outcome while treating talent amplification and client value as the non-negotiable strategic horizon.

The backdrop is an industry-wide arms race. Capital markets and investment banking are data-dense domains where milliseconds, marginal insights and slightly better client service translate directly into fee pools and franchise value. Firms are experimenting with large language models, agentic systems that can orchestrate tasks across multiple internal platforms, and domain-specific models trained on proprietary research and transaction data. Waldron’s framing reflects an awareness that simply swapping humans for machines misses the deeper opportunity: using AI to turn each high-value banker, salesperson, risk manager or wealth adviser into a far more productive and differentiated node in the client network.

One way to read the remark about not knowing what AI means for headcount is as a rejection of simplistic automation narratives. Banks have lived through several waves of technology-induced transformation: the move from floor to electronic trading, the rise of algorithmic execution, the digitisation of retail banking, and the first generation of robotic process automation in operations. Each time, job counts moved in complex ways. Some roles vanished, others morphed, and entirely new categories – from electronic trading quants to cybersecurity teams – emerged. AI promises a wider, deeper transformation, but its labour market effects are still non-linear and hard to predict with precision.

Yet the conviction that the most capable people will become more capable is not just comforting rhetoric. It is grounded in how AI tools actually work at scale inside a financial institution. Generative models and AI agents are particularly good at reducing search costs, summarising complex information, drafting first versions of documents, and running scenario analyses that would previously have taken days. For a high-performing banker or portfolio manager, this does not replace the job; it changes the mix of time spent between low-value tasks and high-value tasks. Instead of manually synthesising dozens of analyst reports, regulatory filings and market moves, they can spend more time on structuring bespoke solutions, negotiating, and reading subtle client signals.

In that sense, AI functions as a force multiplier on human capital. It compresses the information-gathering and analysis cycles while leaving the judgment-laden, relationship-heavy decisions to people. This is particularly true in contexts where the stakes are high and the outcomes are path-dependent: complex M&A transactions, large capital raisings, cross-border restructurings and bespoke hedging strategies. Even the most advanced models struggle with accountability and with understanding the political, cultural and interpersonal nuances that govern whether deals succeed. Senior leaders at firms like Goldman Sachs therefore have an incentive to push AI hardest into the parts of the stack where repeatable pattern recognition and calculation dominate, while reinforcing the human ownership of outcomes.

That is also where the idea of “stickier” client relationships enters the picture. In investment banking and institutional sales, stickiness is not just about price or product; it is about the client’s sense that their adviser understands them better than anyone else and can move faster and more creatively on their behalf. AI, deployed well, can support this in two ways. First, it enhances institutional memory: systems can surface prior deal structures, historical interactions, risk incidents and bespoke constraints instantly, making each interaction feel more tailored. Second, AI agents can coordinate execution across trading, research, legal, risk and operations, reducing friction and error. When a client’s experience shifts from cumbersome and bureaucratic to anticipatory and seamless, they are less inclined to shop around.

However, this potential is contingent on a careful balancing act between automation and authenticity. If clients perceive that their supposedly “trusted” relationship is mediated primarily by bots, templated responses or generic recommendations, trust erodes. Waldron’s emphasis on making capable people more capable implicitly recognises this. The strategic bet is that AI will remain largely invisible at the surface of the relationship, embedded behind the scenes in analytics, workflow orchestration, compliance checks and personalisation engines, while the main interface remains a human team. The more that team is equipped with timely insights, contextual prompts and operational support, the more distinctive their service feels.

Factual context also matters. Goldman Sachs has been investing in technology infrastructure and data platforms for years, building the plumbing needed for more advanced AI deployments. Initiatives in transaction banking, electronic execution, consumer platforms and internal developer tooling have all contributed to a landscape where data is more accessible and systems more modular than in a traditional, siloed bank. Public commentary from the firm’s leadership emphasises not only generative AI but also narrow models embedded in risk management, fraud detection and operational resilience ¹. This context helps explain why the uncertainty is about people rather than about the technology itself: the leadership is betting that the technical foundations are already good enough to support ambitious AI use cases.

There is also an internal cultural dimension. Investment banks are talent-driven organisations, and senior management must signal to their workforce how to feel about AI. If the message were primarily about cost cutting and role elimination, the effect would be corrosive, driving away exactly the kind of high-initiative employees who are most needed to integrate AI into real workflows. By contrast, Waldron’s framing positions AI as a capability enhancer for “really capable, smart people”. This is not just flattery; it is a way of aligning incentives. The people most motivated to experiment with, and adopt, new tools are those who can see a direct line from that adoption to their own performance and client impact.

Yet sceptics will point out that large financial institutions are unlikely to ignore the cost-reduction potential of automation. Operations, middle-office functions, compliance checks, trade processing and certain forms of research are all ripe for productivity gains. Over time, this will almost certainly translate into fewer people doing some categories of work. The unresolved question is whether the new roles created – AI product managers inside banking divisions, data stewards, governance specialists, model risk experts, prompt engineers embedded in deal teams – will be enough to offset the roles displaced. Waldron does not pretend to know the net number, and that humility is itself telling: credible leaders today avoid making precise headcount forecasts about technologies whose adoption curves and regulatory constraints are still in flux.

Strategically, the remark hints at a broader shift from thinking about “jobs” to thinking about “tasks” and “capabilities”. Even if a job title, such as equity research analyst or loan operations specialist, persists, the task composition of that job may change radically. Routine tasks are more exposed to automation; non-routine, interpersonal and judgment-heavy tasks are more likely to be augmented. From a management perspective, the right question is no longer “Which jobs will disappear?” but “Which tasks can we automate so that scarce human attention is reallocated to higher-value activities?” AI agents that can execute multi-step workflows across internal systems accelerate this decomposition of work into modular components.

For example, in a simplified depiction of a workflow, a human banker today might spend a significant share of time assembling data, populating pitch books, and manually checking constraints across risk, legal and tax. With AI agents orchestrating these steps in the background, the banker’s time is reweighted towards designing creative structures, rehearsing negotiation angles, and cultivating the client relationship. The measurable productivity of that banker – deals originated per year, win rate on mandates, revenue per head – could rise substantially. From the firm’s perspective, it becomes rational to concentrate more resources around such high-leverage individuals, further reinforcing the idea that the best people become even more central, not less.

This naturally raises concerns about inequality within the workforce. If AI disproportionately augments those already at the top of the performance distribution, while automating away more routine work, the result could be a more polarised organisation with a smaller middle. Junior staff might worry that the traditional apprenticeship model – learning by doing the grunt work – will erode if AI handles many of the foundational tasks. Leaders need a plan for how to maintain skill development pathways in a world where a model can generate 80 % of the first draft and a set of agents can assemble the data room overnight.

One plausible response is to redesign training and progression so that early-career professionals are explicitly taught how to supervise, critique and improve AI outputs. Instead of spending years perfecting the mechanics of spreadsheet work or drafting, they might move more quickly into roles that require critical thinking, risk awareness and client interaction, with AI handling the routine mechanics under their supervision. This could make the profession more, not less, intellectually demanding from the outset, but it also demands a higher standard of digital literacy and model awareness from new hires. Institutions that manage this transition well will likely find themselves with a more capable and adaptable talent base.

The external regulatory environment adds another layer of complexity. Banks operate under intense scrutiny regarding model risk, fairness, data privacy and operational resilience. Any large-scale use of AI, especially generative models, must be embedded in rigorous governance frameworks: model validation, bias testing, audit trails, human-in-the-loop controls, and clear lines of accountability when things go wrong. This governance overhead can slow down the deployment of AI into certain decision-making processes, keeping humans formally “in charge” in ways that preserve many existing roles. Over time, however, as regulators become more comfortable with well-governed AI systems, the frontier between human and machine decision-rights may shift, with corresponding effects on headcount.

There is also a competitive dynamic among clients themselves. Many of Goldman Sachs’s corporate and institutional clients are deploying AI aggressively inside their own organisations. They will expect their financial counterparties and advisers to understand AI-driven business models, valuation dynamics, and operational risks. For a banker, salesperson or risk adviser, being “more capable” means not only using AI tools internally but also engaging credibly with clients on their AI strategy. This creates a feedback loop: the more AI reshapes the real economy, the more valuable are advisers who can straddle both finance and technology, and the more sense it makes to augment those advisers with powerful internal AI systems.

Critics might argue that narratives centred on augmenting smart people risk obscuring legitimate concerns about transparency and accountability. If decision-making becomes a hybrid of human judgement and opaque model outputs, it can be harder for outsiders – clients, regulators, even boards – to know who is responsible. The same tools that make client relationships stickier can also increase information asymmetries, as banks deploy proprietary AI to extract insights from data that clients cannot see. Addressing these concerns will require a mix of explainable AI practices, transparent governance and, importantly, a culture that treats AI as an advisory input rather than an oracular verdict.

Another objection is that emphasising “stickier” relationships and stronger value propositions could be read as a soft way of saying “higher margins” and “greater pricing power”. Clients may worry that AI allows banks to segment them more finely, extract more surplus, or cross-sell more aggressively. There is some truth to this: better analytics and personalisation do sharpen commercial strategies. But the counterpoint is that sophisticated clients are not passive; they can benchmark, negotiate and multi-home across providers. In this environment, the institutions that use AI to genuinely improve outcomes – better execution quality, more resilient portfolios, more innovative deal structures – will likely be rewarded with loyalty, while those that use it primarily to squeeze clients may face backlash.

What makes Waldron’s comments distinctive is the explicit refusal to present AI as a neat, linear story about job counts. Instead, the focus is on a direction of travel: towards a firm where the human core remains central but is surrounded by increasingly capable digital infrastructure, and where client relationships are deepened through better insight and execution rather than merely defended by incumbency. The underlying bet is that in a market where many players will have access to similar foundational models, the differentiator will not be the model itself but the combination of data, culture, governance and human talent wrapped around it.

For Goldman Sachs, and for its peers, the stakes are high. If they over-automate, they risk hollowing out the very human capabilities that justify premium fees and enduring relationships. If they under-invest, they risk being outcompeted by more agile players, including technology firms encroaching on financial intermediation. Steering between these extremes requires exactly the kind of nuanced view reflected in Waldron’s statement: technological confidence paired with organisational humility, and a willingness to let the precise headcount outcomes emerge from a series of experiments rather than from a predetermined spreadsheet.

Ultimately, what matters is not whether the aggregate number of jobs goes up or down by some percentage, but whether the institution manages to re-architect work so that human potential is amplified rather than diminished. If AI makes the best people better, gives more clients access to that enhanced capability, and does so under robust governance, then the technology will have served as a catalyst for a new kind of financial advisory practice. If, instead, it leads to shallow automation, brittle systems and eroded trust, the promise of stickier relationships and stronger value propositions will remain unfulfilled. Waldron’s remarks implicitly challenge his organisation – and, by extension, the industry – to aim for the former path, even while admitting that the journey’s precise employment contours cannot yet be mapped.