Enterprises Stopped Buying AI. They Started Hiring It.

AI does not install the way SaaS did, and once it is running, it does not supervise itself. We tracked Forward Deployed Engineer and Agent Manager job posting volume from 2021 to 2026. The two roles outgrew every other enterprise AI title in 18 months.

SaaS won the last two decades of enterprise software by removing the implementation tax. Customers paid for seats and got hosted access. AI broke that model on both ends. Every enterprise AI deployment requires custom engineering inside the customer's environment, because the model only reaches usable performance after integration with the customer's data and identity stack. What the deployment produces behaves more like a team member than a tool. Agents need supervision and ongoing retraining the way a junior employee would.

Forward Deployed Engineer postings grew from a Palantir-only category in 2021 to over 5,300 active US postings in April 2026 alone. Agent Manager and its precursor titles grew alongside them, with Salesforce, Coinbase, and ServiceNow formalizing the title in early 2026. Together, the two roles account for an estimated 200,000+ open US positions in 2026, none of which existed in this form five years ago. AI is the first enterprise software category in twenty years that hires more humans into customer-facing operational work each year, not fewer. The hiring data is the proof.

Enterprise software for two decades was a license-and-rollout business. The vendor hosted the product, the customer paid for seats, and IT pointed the configuration screen at whatever already existed. AI does not work that way. Every meaningful enterprise AI deployment in 2024 and 2025 required custom engineering inside the customer's environment, and the systems those deployments produced required ongoing supervision once they reached production. This is the other side of the trend we documented in the quiet contraction of SaaS sales and marketing teams: as legacy GTM headcount thins, a new layer of AI-native operational roles is forming on top.

The two roles that grew fastest in 2025 each answer one of those problems. Forward Deployed Engineer is the answer to the install problem. Agent Manager is the answer to the supervision problem. Both grew at rates no other AI role came close to.

01The Two Roles, Defined

FDE is an engineering role that ships software inside customer environments. SaaS-style configuration UIs cannot do the work between the API and the customer outcome: identity setup, retrieval pipeline configuration, evaluation harnesses, guardrails, and integration with whatever CRM or data warehouse the customer already runs. FDEs at OpenAI and Anthropic write that code on customer premises. Anthropic's lead software engineer compensation tops out at $785K on Levels.fyi, which gives a sense of what frontier labs are willing to pay for the role.

Agent Manager is an operational role, and the job has no SaaS-era precedent. Once an enterprise runs ten or more production agents, no single product owner can audit them. The work covers observability across an agent fleet, escalation policy when an agent hits an edge case it cannot handle, retraining workflows when performance degrades, and capacity planning across human and digital workers. The HBR February 2026 case study of Salesforce's agent manager Zach Stauber described the job as starting and ending the day inside dashboards and agent observability monitoring. The closest analog is a customer success manager who happens to manage software instead of accounts.

The split matters because the two roles report to different orgs. FDE sits in Engineering. Agent Manager sits in RevOps, Customer Success, or whichever business unit owns the agents themselves. Treating them as one role is the most common mistake we see in operator hiring plans.

02The Demand Curves Diverge in Shape

FDE postings sat flat from 2021 to 2023 because the title was effectively Palantir-only. Adoption inflected in late 2023 when OpenAI hired its first two FDEs. By the end of 2024, OpenAI's FDE team had grown to 39 engineers under Colin Jarvis, with plans to reach 52. Bloomberry's analysis of 1,000 FDE postings showed 1,165% year-over-year growth between January–October 2024 and January–October 2025. Indeed data shows another 729% expansion from April 2025 to April 2026.

Agent Manager's growth curve looks different. Pre-2024, the role existed as Automation Manager or RPA Operations Manager, anchored to UiPath and similar RPA platforms. Growth was steady at 5 to 15% per year. The inflection came with agentic AI itself. Stanford's AI Index 2025 reported agentic-AI skill postings grew 986% between 2023 and 2024. The titles caught up in 2025: Salesforce introduced formal Agent Manager roles inside Agentforce, ServiceNow followed with AI Operations Manager, and Coinbase referenced “AI Agent Manager” on a May 2026 earnings call.

The chart shows two stories on one canvas. FDE grew as a step function: Palantir for most of the decade, then a cluster of frontier labs and Series A startups all at once, with the inflection between 2023 and 2024. Agent Manager grew as a renaming wave, with existing automation and RPA roles repositioning before new roles got created at scale. The 2026 trajectory for both is set. The operator question is which one to hire first.

03Where the Demand Concentrates

Both roles cluster in SaaS and applied AI. The split below that line is where the operator decision lives. FDE postings concentrate in Financial Services at 27%, driven by data complexity and compliance overhead inside banks and fintechs. Consulting and IT services account for 18%. Agent Manager flips the GTM share: Sales and GTM operations account for 20% of Agent Manager postings versus 8% of FDE postings, because the first wave of production agents in 2025 ran inside CRMs and sales engagement tools. When agents live in sales workflows, the operational owner reports to RevOps, not IT.

Financial Services trails SaaS for Agent Manager because regulated industries adopted agents slower. That gap closes in 2026 as Coinbase, Stripe, and the larger fintechs formalize agent operations as a function.

04Who Is Actually Hiring

The company-size distribution explains why FDE and Agent Manager are not the same job. FDE postings concentrate in mid-market companies: 47% of FDE roles sit at organizations between 51 and 1,000 employees, mostly Series A through Series C AI-native startups. Paraform's 2026 marketplace data showed 59% of FDE-hiring companies were Seed through Series A. The pattern tracks the broader rotation in what Series A capital now funds, where AI-native categories absorbed the share generalist software lost.

Agent Manager skews enterprise. 66% of Agent Manager postings sit at companies with over 1,000 employees. The role only makes economic sense when an organization runs enough production agents that no single product owner can supervise them. The threshold sits around 10 concurrent agents in production. Below that, the work gets absorbed by existing product or RevOps managers.

The implication for founders is direct. If the company has 200 employees and three customers, the right hire is an FDE, not an Agent Manager. Reverse the headcount and the answer reverses too.

05The Geography Splits Differently

The geography of FDE and Agent Manager hiring is not the same as the geography of AI research. NYC leads FDE hiring with 24% of global openings, driven by fintech and regulated-industry demand. The Bay Area sits at 22%, anchored by frontier labs. India's GCCs account for 14% of FDE roles, mostly delivery-scale work inside Accenture, EY, and McKinsey QuantumBlack centers.

Agent Manager hiring concentrates differently. India GCCs lead at 26% of global Agent Manager openings because the role is operationally heavy and scales through 24/7 coverage. The Bay Area sits at 19%, NYC at 14%. EY's 2025 GCC Pulse Survey found 58% of Indian GCCs were already investing in agentic AI, and 83% were scaling generative AI.

London is the second-tier hub for both roles, anchored by Anthropic's 158,000 sq ft One Triton Square lease and OpenAI's 88,500 sq ft King's Cross HQ. The radar shape shows what a table cannot: FDE is a US product-hub story, Agent Manager is a global operations story.

06The Compensation Premium Is Widening

Base salaries for both roles rose every year from 2021 to 2026, but the slopes are not equal. FDE average base in 2021 was around $135K, anchored to Palantir FDSE comp. By 2026, the average sits at $214K, with frontier-lab packages running far higher. Anthropic's lead software engineer total compensation reaches $785K on Levels.fyi; OpenAI's median software engineer total comp is $555K; Stripe's professional services FDE band runs $173K to $259.6K base.

Agent Manager base salary grew from $98K in 2021 (the Automation Manager and RPA Operations Manager era) to $152K in 2026. Salesforce's Agentforce/AI Deployment Strategist postings run $184K to $351.8K base in New York and San Francisco. The premium is real, but the spread is narrower than FDE.

The gap between the two roles is widening, not closing. FDE compensation grew 58% over the five years; Agent Manager grew 55%. The difference is the ceiling. An FDE at a frontier lab can clear $500K total compensation in year one. An Agent Manager at a Fortune 500 with five years of agent operations experience tops out around $350K. The reason is supply: engineers who can ship production AI inside customer environments are scarce, while operational managers who can run agent fleets are growing faster, partly because the role can be filled from existing RevOps and customer success ranks. The same scarcity logic shows up in our lean sales efficiency data, where the highest revenue per employee comes from small, AI-augmented teams rather than large ones.

07Conclusion

Enterprise AI in 2025 produced two new categories of expensive labor, and the production of those categories is itself the punchline. SaaS removed the human implementation layer for two decades. AI reintroduced it on both ends. The Forward Deployed Engineer exists because the install problem came back. The Agent Manager exists because the supervision problem was new to enterprise software entirely. Both roles grew from negligible 2021 baselines to combined estimated US posting volume of over 200,000 per year.

MIT Project NANDA's 95% failure rate for enterprise GenAI pilots and Gartner's 40% cancellation forecast for agentic AI projects both point to the same conclusion: the bottleneck is operations, not the model layer. The operators who treat these roles as nice-to-have join the failure cohort. The ones who staff them early compress deployment timelines from quarters to weeks and produce production systems that the rest of the market still calls pilots.

The framework is straightforward. If the company is shipping AI into customer environments, hire FDEs first. If the company is running AI inside its own workflows, hire Agent Managers first. If it is doing both, it needs both, and the headcount math should reflect that ratio. AI did not arrive as a product to buy. It arrived as a workforce to hire, and the org chart is catching up.

08Implications · Q&A

The data points to concrete takeaways for operators deciding which role to staff, when, and where.