When Mohamad Ali, Senior Vice President (SVP) and Head of IBM Consulting, published his blog post yesterday, the core message was unmistakable: the traditional consulting model no longer works. “Services as Software” – that is what Ali calls the future. Consulting is becoming a combination of people and software, delivered through intelligent agents and platforms that scale far beyond traditional project work. IBM itself has redesigned over 100 workflows using artificial intelligence (AI) since 2023 and reports USD 4.5 billion in productivity gains. The message to the industry is clear: those who cling to the old model will become irrelevant.
Yet the matter is not that simple. Whilst Ali proclaims the end of the old consulting world, the industry is experiencing perhaps its greatest paradox: revenues are rising – and jobs are shrinking. BCG reports record revenues of USD 13.5 billion, of which USD 2.7 billion came from AI consulting alone. Accenture has booked USD 2 billion in generative AI engagements. At the same time, the number of entry-level consultants has been declining measurably since the 2023 peak. McKinsey has cut approximately 5,000 positions over five years. Overall demand for traditional consultants stands 40 per cent below its 2023 high.
What is happening here is not a contradiction – it is a selection. Consulting is not dying. But a particular model of consulting is. And something fundamentally new is taking its place.
The Slide Factory Has Had Its Day
Luk Smeyers, founder of The Visible Authority and former partner at Deloitte Analytics, puts it succinctly: “Real consulting begins where playbooks end.” What is disappearing is what he calls the “slide-heavy, junior-stacked model” – the consulting factory where graduates translate data into PowerPoint decks that partners then present to clients.
The Harvard Business Review (HBR) analysed this structural shift in a widely noted September 2025 article and describes an “Obelisk Model”: where a broad base of junior analysts once supported the pyramid, a leaner structure with three new core roles is now emerging. AI Facilitators design AI workflows. Engagement Architects define problems and interpret AI outputs. Client Leaders maintain C-suite relationships and steer strategic direction. The analyst base is shrinking – AI is taking over their work.
–10% entry-level consultant headcount since the 2023 peak (Revelio Labs)
–40% overall demand for consultants versus the 2023 high
~11,000 positions cut at Accenture in AI-driven restructuring
72% of McKinsey’s 40,000 employees use the internal AI platform Lilli
36.2% CAGR projected growth of the AI consulting market (2024–2030)
The major firms are responding with massive investments: Accenture is putting USD 3 billion into its Data & AI Practice. Deloitte is investing USD 2 billion in its “Industry Advantage” programme. PwC has become OpenAI’s largest enterprise customer. KPMG has signed a USD 2 billion alliance with Microsoft. EY is building its own Large Language Model (LLM) platform called EY.ai. McKinsey, BCG, Accenture and Capgemini have been certified partners of the OpenAI Frontier Alliance since February 2026 – implementing the enterprise agent stack directly at client sites.
The industry is investing billions in its own transformation. But who exactly is supposed to deliver this transformation at the client’s end? This is where a role enters the picture that, two years ago, hardly anyone outside Palantir had heard of.
Enter the Forward-Deployed Engineer
The term Forward-Deployed Engineer (FDE) has military origins and was adapted by Palantir for the software industry in the early 2010s. Internally, Palantir distinguished between “Dev” (one capability, many customers) and “Delta” (one customer, many capabilities). The FDEs were the Deltas – engineers stationed directly at client sites, not to advise but to build. By 2016, Palantir had more FDEs than traditional software engineers.
What distinguishes the FDE from the traditional consultant is not proximity to the client – both share that. It is the output. The consultant delivers recommendations. The FDE delivers working code. The consultant hands over a slide deck. The FDE debugs at two in the morning. The consultant closes their project. The FDE is embedded for three to twelve months, and their insights flow back into the product roadmap.
Marty Cagan of the Silicon Valley Product Group (SVPG) contextualises this strategically: Palantir applies the product model to custom solutions. Unlike consulting firms that build what the client specifies, Palantir takes ownership of outcomes. This only works when engineers have direct access to clients, data and stakeholders. Cagan points to valuations as evidence: as of September 2025, Palantir’s market capitalisation exceeded USD 400 billion versus Accenture’s USD 150 billion – with a fraction of the headcount. The precise figures fluctuate with the market, but the structural point remains: outcome ownership is valued more highly than delivery capacity.
The Hottest Job in Tech
What was long a niche phenomenon exploded in 2025. Andreessen Horowitz (a16z) called the FDE “the hottest job in startups” in June 2025. Demand is significant: on OpenAI’s careers page alone, 22 of 311 open positions at the time fell into the FDE category. Only around 1 per cent of all companies currently have an FDE role – but the curve is steep, and the number of job postings multiplied over the course of 2025 according to industry analyses.
OpenAI is building its own FDE team – currently over ten engineers, with around 50 expected by year-end. Databricks, Anthropic and Cohere have FDEs. Ramp, the fintech company, operates with 15 FDEs in small pods. Salesforce is hiring FDEs to implement Agentforce. And in December 2025, Deloitte made it official: “Deloitte Forward Deployed Engineering” – the Big Four are adopting the model directly.
The Triad: Consulting + Architecture + Coding
What makes the FDE truly interesting is not the fact that they write code. It is the hybrid skill mix the role demands. Amit Koth describes it aptly: “The role is a hybrid – part software engineer, part product manager, part consultant. Someone comfortable writing production-quality code while also explaining to the CFO why this automation will cut costs.”
On any given day, an FDE might work as a data engineer integrating a legacy database in the morning, fix a UI bug as a frontend developer at midday, and spend the afternoon discussing process optimisation with department heads as a business analyst. This is not generalism – it is the triad of consulting competence, solution architecture and programming that the new reality demands.
The Variable Weighting
The crucial point: this triad is not a fixed formula. The weighting shifts – depending on context, client and project phase.
| Context | Consulting | Architecture | Coding |
|---|---|---|---|
| Palantir (Government/Defence) | Medium | High | High |
| OpenAI FDE | Low | High | Very high |
| Deloitte FDE | High | Medium | Medium |
| Ramp / Fintech | Low | Medium | High |
| Financial Services Transformation | High | High | Medium |
Colin Jarvis, Head of FDE at OpenAI, also describes the shift across project phases: in the scoping phase, the consulting element dominates – stakeholder workshops, process mapping, prioritisation. In the validation phase, the emphasis shifts to architecture and coding – building evaluations, developing features. In the delivery phase, it is primarily coding – integration, debugging, production operations.
For the financial industry – regulated, complex, stakeholder-intensive – the consulting component is naturally higher than, say, at a fintech startup. Anyone implementing an AI-powered regulatory reporting platform must not only be able to write code but also understand the regulatory requirements of the Federal Financial Supervisory Authority (BaFin), navigate the client’s organisational structure and convince the business side of the technology. That is consulting, architecture and coding in one – with adjustable sliders.
Why the FDE Is No Silver Bullet
The counter-argument deserves attention – and it is substantial. The FDE model has structural limits that its advocates often underestimate.
The fundamental problem is scalability. One client, one FDE – that works. Ten clients, ten FDEs – manageable. Ten thousand clients? The model breaks down. AI implementation complexity scales non-linearly: every environment has unique data structures, legacy systems, workflows and compliance requirements. There is no universal integration template. “AI is still artisanal,” observes FourWeekMBA.
Flybridge Capital estimates – based on conversations with former Palantir FDEs – that 95 per cent of startups misapply the FDE model. It only works with complex workflows, high contract values with a clear path to seven-figure deals and a strong modular platform. Deploying an FDE on a client with less than USD 100,000 in annual contract value burns money – a fully loaded FDE costs between USD 220,000 and USD 400,000 per year.
And then there is organisational blindness. AI cannot read internal corporate reality – power dynamics, informal coalitions, the real question behind the stated question. Bill Conerly argues in Forbes, citing Thomas Sowell: “There are no solutions, only tradeoffs.” Critical business decisions involve human value judgements that AI can neither prioritise nor weigh. Security investments beyond regulatory minimums, data protection beyond pure cost calculations, corporate culture in workforce decisions – all of this remains human.
The paradox deserves emphasis: FDE teams at OpenAI, Anthropic and Cohere grew three- to fivefold in 2024 and 2025 – contrary to the “self-service soon” thesis. We are in Stage 2: humans are teaching AI how organisations work. Stage 3 – AI adapting on its own – has not been reached.
Long Live Consulting
So: consulting is dead? No. A particular model of consulting is dead – the slide-factory model built on junior arbitrage. What lives and grows is judgement-intensive, stakeholder-heavy, context-sensitive consulting in transformations, crises and regulated environments.
Executive alignment when leadership teams are divided during critical transformations. Change management when new ways of working must be implemented across diverse teams. Post-merger integrations where multi-year ERP programmes cannot afford to fail. Crisis advisory under time pressure with incomplete information. Compliance in sectors where contextual heterogeneity defeats every one-size-fits-all solution.
The Global Consultants Review puts it succinctly: “AI solves ‘What’ and ‘How’ – not ‘Why’.” AI does not understand why an organisation behaves the way it does: internal politics, cultural resistance, leadership egos, unspoken decision constraints. And trust – the true currency of consulting – cannot be automated. It emerges when consultants listen, ask uncomfortable questions and take responsibility for their recommendations.
What is changing is the delivery model. The modern consultant – whether they call themselves an FDE, Engagement Architect or something else entirely – must be able to do three things simultaneously: understand the client’s business problem (consulting), design the technical architecture of the solution (solution architecture) and build the solution themselves or at least competently steer it (coding). The weighting is variable. But none of the three dimensions is optional.
IBM chief Ali is right when he says: “Firms that hold on to traditional delivery models will struggle.” But he omits the second part, which is equally true: firms that believe technology alone can replace human judgement in complex organisations will fail just as surely.
The future belongs to those who can do both. It is time to reinvent yourself.
Recommendations for Action
What consultants, consulting firms and their clients should prioritise now – regardless of whether they come from traditional consulting or technology.
Where do I stand in the triad of Consulting + Architecture + Coding? Technical depth is the credibility mechanism of the new consulting world – anyone who does not understand production-ready code will lose conversational authority with the client within three months at most.
The majority of the industry still bills by the hour. Those who switch to outcome-based models earlier will win the trust of clients who no longer want to pay for slide decks – but for working solutions.
IBM’s argument is valid: if you operate 3,000 internal agents, you can credibly advise clients on their implementation. “Eat your own dog food” has never been more relevant.
An FDE at a financial institution is not the same as an FDE at a fintech startup. The consulting component in regulated industries is higher – and that is not a deficit but the value that experience and judgement create.
Technology automates analysis, research and pattern recognition. It does not automate the conversation with a CEO facing a decision under incomplete information. This human dimension endures – and is becoming more valuable, not less.
Timeline: The Transformation of Consulting
Keep reading – in your inbox every two weeks.
Capital markets insights, regulatory updates and AI trends. Concise, well-founded, free of charge.
GDPR-compliant. Unsubscribe at any time.