Enterprise AI needs hands-on work
Companies deploying AI agents at scale are running into a manpower problem: they can't hire enough forward-deployed engineers to implement and customize agents, and customer‑success teams are struggling to fill the gap. OpenAI's documentation also frames enterprise use in operational terms — routing, planning and admin oversight matter more than raw model claims — showing deployment is now a configuration and integration task rather than a plug‑and‑play demo. (saastr.com) (help.openai.com)
Enterprise AI projects are turning into staffing projects, because companies still need people to wire agents into real systems. (saastr.com) SaaStr founder Jason Lemkin wrote this week that teams rolling out agents at scale “can’t hire enough forward deployed engineers,” the technical staff who implement, tune, and customize software inside customer environments. He also argued customer-success teams usually cannot absorb that work because the job is closer to engineering than account management. (saastr.com 1) (saastr.com 2) OpenAI’s own enterprise material describes the same problem in operational terms. Its February 5, 2026 launch post for Frontier said what slows companies down is not model intelligence but “how agents are built and run,” and the product pitch centered on shared context, onboarding, permissions, feedback, and governance. (openai.com) That is a shift from the chatbot era, when vendors could sell a demo. OpenAI’s agent-building guide says an agent is a system that manages a workflow, uses tools to reach outside systems, and hands control back when it fails, which makes deployment look more like process design than prompt writing. (openai.com) The implementation burden shows up in the product docs. OpenAI’s Agents documentation tells builders to define instructions, tools, handoffs, guardrails, approvals, and structured outputs, and to use dynamic instructions when behavior depends on a customer or runtime context. (developers.openai.com) The administrative layer is expanding too. OpenAI updated ChatGPT Enterprise on April 2, 2026 with a new Codex-only seat type and token-based pricing for new Enterprise customers, adding another set of roles, access controls, and usage rules for companies to manage. (help.openai.com 1) (help.openai.com 2) OpenAI said on April 8 that enterprise now makes up more than 40% of its revenue and is on track to reach parity with consumer by the end of 2026. In the same note, Chief Revenue Officer Denise Dresser said customers are tired of disconnected AI point solutions and want agents tied to company context, internal systems, and permissions. (openai.com) SaaStr’s recent posts put a price on the labor behind that shift. Lemkin wrote that in today’s market it can cost $50,000 to $70,000 to get one agent started, plus another $25,000 for a forward-deployed engineer to set it up, and argued that better implementation support can beat a better product that never gets configured properly. (saastr.com) The bottleneck is no longer whether an agent can answer a question in a demo. It is whether a company can supply the engineers, approvals, and system access needed to make that agent work inside a live business. (openai.com) (saastr.com)
