Salesforce has stopped hiring engineers: what the AI agent wave means for SMEs
Table of Contents
- Why did Salesforce stop hiring engineers in 2026?
- What does the McKinsey study reveal about real AI agent adoption in business?
- Oracle launches 22 agentic applications: what does this change for businesses?
- How does the French-speaking market position itself in the AI agent race?
- What are the concrete use cases for AI agents in an SME?
- What budget should you plan for deploying AI agents in an SME?
- Persistent memory: the next technological leap for AI agents
- How to start with AI agents in your business
- FAQ
On March 24, 2026, Marc Benioff, CEO of Salesforce, confirmed that his company had hired no new engineers during fiscal year 2026. The reason: internal AI coding agents had driven productivity so high that new hires became unnecessary. The same day, Oracle launched 22 Fusion agentic applications, each powered by coordinated teams of specialized AI agents. Two days earlier, Forbes had reported a McKinsey study showing that only 10% of business functions use AI agents today. These three signals paint a clear picture: large enterprises are accelerating, most SMEs remain on the sidelines, and the window to catch up is closing fast.
For small and mid-sized businesses across France, Switzerland, and the UAE, this is not a technology watch item. It is an operational signal. Companies deploying AI agents now are gaining ground over competitors in terms of cost, execution speed, and processing capacity. Those that wait risk repeating the mistakes of businesses that ignored the internet in 2000 or mobile in 2010. AI automation is no longer a comfort option. It is an economic survival parameter.
Why did Salesforce stop hiring engineers in 2026?
Salesforce stopped hiring engineers because its internal AI coding agents now generate more value than traditional recruiting could deliver.
Marc Benioff stated this on March 24, 2026: the autonomous coding tools used internally produced a productivity jump large enough that the existing workforce, augmented by agents, is sufficient to maintain and accelerate product development. Salesforce employs approximately 72,000 people worldwide (source: FY2025 annual report). The fact that a company of that size can freeze technical hiring without slowing its roadmap illustrates the power of autonomous AI agents in software development workflows.
This is not an isolated case. Andrej Karpathy, former AI director at Tesla, stated in March 2026 that coding agents now write 80% of his code. Cognition, the startup behind Devin, raised $175 million in 2025 on the promise of an autonomous developer agent. GitHub Copilot already processes more than 150 million suggested lines of code per day worldwide (source: GitHub, February 2026). The trend is systemic: the developer role is shifting from code producer to code agent supervisor.
| Company | AI agent decision | Date | Source |
|---|---|---|---|
| Salesforce | Zero engineer hires FY2026 | March 2026 | Latestly, Benioff |
| NVIDIA | 100 agents per employee targeted | March 2026 | GTC 2026 |
| Oracle | 22 Fusion agentic applications | March 24, 2026 | Oracle AI World |
| Alibaba | Wukong enterprise agent platform | March 2026 | CNBC |
| Microsoft | Azure Copilot Migration Agent | March 2026 | AI Business |
What does the McKinsey study reveal about real AI agent adoption in business?
McKinsey reveals that only 10% of business functions use AI agents today, meaning 90% of the potential remains untapped.
The study, reported by Forbes on March 22, 2026, shows a massive gap between the ambient discourse around AI and operational reality. Aaron Levie, CEO of Box, commented on these figures by noting that the real challenge is not the technology but the integration into existing processes. According to CB Insights, AI startups absorbed more than half of all global venture capital in the first half of 2025, an unprecedented influx of capital into the sector.
For SMEs, the 10% figure is an opportunity. It means the majority of companies, including large enterprises, have not yet made the move. A business that deploys AI agents on its core functions (accounting, customer relations, logistics, lead qualification) can build a meaningful lead over local competitors. The entry cost has dropped sharply: a specialized AI agent costs between 200 and 2,000 euros per month in tokens, compared to 4,000 to 8,000 euros per month for a full-time employee performing the same function.
| AI agent adoption indicator | Value | Source |
|---|---|---|
| Business functions using AI agents | 10% | McKinsey / Forbes, March 2026 |
| Companies planning a 2026 deployment | 42% | Gartner CIO Survey 2026 |
| Fortune 500 with an active AI agent project | 67% | Deloitte, Q1 2026 |
| Share of global VC captured by AI startups (H1 2025) | More than 50% | CB Insights |
| AI agent market growth 2025-2028 | +47% CAGR | MarketsandMarkets |
Oracle launches 22 agentic applications: what does this change for businesses?
Oracle is transforming its cloud applications into systems driven by coordinated teams of AI agents, capable of reasoning, deciding, and acting without constant human intervention.
Announced on March 24, 2026 at Oracle AI World in London, the Fusion Agentic Applications launch covers 22 business functions: finance, HR, supply chain, sales, and customer service. Each application relies on teams of specialized agents that divide tasks among themselves. One agent analyzes data, another formulates recommendations, a third executes the action in the system. The entire process is supervised by deterministic guardrails that prevent unauthorized actions.
For SMEs, the signal is twofold. First, major software vendors are embedding agents directly into their business applications, which will democratize access. Second, companies that wait for their ERP to be updated before discovering agents will lose time. Those that build custom agents now, adapted to their specific processes, will hold a structural advantage. Oracle targets large accounts with Fusion, but the logic applies equally to SMEs: decompose each process into micro-tasks, assign each micro-task to a dedicated agent, and supervise the whole with clear rules.
How does the French-speaking market position itself in the AI agent race?
France is advancing in research and startups, but lags behind the United States and China in enterprise deployment.
On the positive side: Mistral AI launched Forge in March 2026, a service for building custom AI models for businesses and governments (source: L'Usine Digitale). Parallel, a French startup, raised $20 million to deploy AI agents in hospitals, with 40 facilities already equipped (source: L'Usine Digitale, March 2026). Oracle France presented its Fusion AI agents at its AI World Tour in Paris (source: Le Monde Informatique). The French football team is using Google Gemini as its official AI assistant for the 2026 World Cup (source: Le Big Data).
On the negative side: according to the France Num 2025 barometer, only 5% of French SMEs use AI in their daily processes. The Thales Data Threat 2026 report, covering 20 countries, highlights that agentic AI is creating a new risk regime that most companies do not know how to manage (source: IT Social). And three studies published in March 2026 converge on the same finding: organizations know they are unprepared, but they are accelerating anyway under competitive pressure (source: IT Social).
| Market signal | Detail | Source |
|---|---|---|
| Mistral AI Forge | Custom AI models for enterprises | L'Usine Digitale, March 2026 |
| Parallel | $20M raised, 40 hospitals equipped with AI agents | L'Usine Digitale, March 2026 |
| SMEs using AI | Only 5% | France Num 2025 barometer |
| Thales Data Threat 2026 | Agentic AI = new risk regime | IT Social |
| Oracle (France, UAE, Switzerland) | Fusion agents presented at local events | Le Monde Informatique |
What are the concrete use cases for AI agents in an SME?
The highest-ROI use cases for an SME are automated lead qualification, accounting document processing, first-level customer support, and competitive intelligence.
A lead qualification agent analyzes incoming requests (forms, emails, calls), ranks them by potential, and prepares a context sheet for the salesperson. Measured result: 60 to 70% reduction in manual qualification time (source: HubSpot State of AI 2025). An accounting agent extracts data from invoices, reconciles them against purchase orders, and prepares journal entries. Result: 15 to 20 hours saved per month for a 50-person company (source: sector estimates, Sage 2025). A customer support agent answers level-1 questions (hours, order tracking, FAQ), escalates complex cases, and drafts summaries. Result: 40 to 60% of tickets resolved without human intervention (source: Zendesk AI Report 2025).
These use cases share one thing: they do not replace employees, they eliminate low-value tasks to free up time for strategic activities. A salesperson who spends 3 hours a day qualifying leads can redirect that time to negotiation and closing. An accountant who spends 4 hours a day on data entry can focus on financial analysis and advising the CEO.
What budget should you plan for deploying AI agents in an SME?
The deployment budget for an SME of 10 to 100 employees ranges from 5,000 to 30,000 euros for the initial phase, then 500 to 3,000 euros per month in ongoing operations.
The initial phase covers process auditing (identifying automatable tasks), agent design (defining rules, data sources, guardrails), and technical deployment (connecting to existing tools: CRM, ERP, messaging). This phase takes 2 to 8 weeks depending on complexity. The main cost is engineering, not technology: language models cost between 1 and 15 euros per million tokens, which represents thousands of pages of processed text.
In operation, the monthly cost depends on the volume of tasks processed. A lead qualification agent that handles 500 requests per month consumes between 100 and 300 euros in tokens. An accounting agent that processes 200 invoices per month consumes between 50 and 150 euros. The cost-to-benefit ratio is compelling: an agent at 200 euros per month replaces 15 to 20 hours of manual work, equivalent to a part-time role at minimum wage. The investment pays back in 2 to 4 months in the majority of cases.
Persistent memory: the next technological leap for AI agents
AI agents in 2026 still suffer from a major limitation: they forget everything between sessions. Persistent memory is the technical barrier currently being resolved.
Several frameworks are emerging to address this. Mem0 provides a memory API compatible with 24 vector database backends. Letta builds a tiered memory directly into its agent framework. Memoria, presented in March 2026, proposes a Git-inspired memory versioning system (source: DEV Community). DeepLearning.AI launched a dedicated course on building agents with persistent memory (source: DeepLearning.AI, March 2026). The issue is critical: an agent without memory repeats the same mistakes, asks the same questions, and loses accumulated context. An agent with memory learns from past interactions, refines its responses, and becomes more effective over time.
For SMEs, persistent memory transforms a disposable tool into a durable collaborator. A customer support agent that remembers past interactions with a client delivers a personalized experience impossible to reproduce manually. A sales agent that retains each prospect's preferences prepares more relevant proposals. Memory is what turns a tool into a strategic asset.
How to start with AI agents in your business
The most effective approach is to identify a single high-volume repetitive task, deploy an agent on it in 2 to 4 weeks, measure results, and then expand progressively.
Step 1: map tasks. List all repetitive tasks in your business with their frequency and duration. Prioritize those that consume the most human time for the least added value. Step 2: choose the first agent. Select a precise, measurable task with sufficient volume (at least 50 occurrences per month). Step 3: deploy and measure. Configure the agent, run it for 2 weeks in parallel with the existing process, then compare results. Step 4: expand. If the first agent proves its worth, identify the next task and repeat.
Mistakes to avoid: trying to automate everything at once, choosing a use case that is too complex to start with, neglecting guardrails (human validation on critical actions), and forgetting team training. An AI agent is not software you install and forget. It is a digital collaborator that requires initial supervision, rule adjustments, and a progressive learning curve for the team managing it.
Orchestra Intelligence helps SMEs and mid-sized companies deploy custom AI agents, from process auditing to production. Our methodology is built on specialized agents, transparent token-based pricing (10 to 20 times cheaper than traditional solutions), and human support at every step. We also offer AI training to upskill your teams and a free diagnostic to identify your first use cases. Learn how companies are already deploying AI agents in production.
FAQ
Can an AI agent really replace an employee in an SME?
No, an AI agent does not replace an employee. It replaces the repetitive, low-value tasks that the employee performs. The employee is freed to focus on activities that require judgment, creativity, and human relationship. McKinsey estimates that 60 to 70% of business tasks are partially automatable, but only 5% of jobs are fully replaceable by AI (source: McKinsey Global Institute).
How long does it take to deploy a first AI agent?
Deploying a first AI agent takes between 2 and 8 weeks depending on the complexity of the use case. A simple lead qualification agent can be operational in 2 weeks. An accounting agent connected to an ERP requires 4 to 6 weeks. A multichannel agent (email, phone, chat) takes 6 to 8 weeks. The longest phase is auditing processes and defining rules, not technical development.
Are AI agents GDPR compliant?
Yes, provided the fundamental principles are respected: data minimization, lawful basis for processing, rights of access and deletion, and hosting of data in Europe. AI agents deployed on European infrastructure (OVH, Scaleway, Supabase EU) and configured with compliant data retention rules are fully compatible with GDPR. The EU AI Act, in force since 2025, imposes additional obligations for high-risk systems.
What is the difference between a chatbot and an AI agent?
A chatbot answers questions based on predefined scripts or a language model. An AI agent reasons, plans, and acts autonomously. It can chain multiple steps, use external tools (APIs, databases, files), make decisions, and adapt to context. Alibaba draws a clear distinction between the two in its Wukong platform: chatbots respond to prompts, agents take proactive actions (source: CNBC, March 2026).
Where do I start if I know nothing about AI?
Start by identifying your three most time-consuming and repetitive tasks. Then have a specialist run a diagnostic that evaluates the automation potential of each task, the budget required, and the expected return on investment. Do not start with the technology, start with the business problem. Orchestra Intelligence offers a free AI diagnostic for SMEs that want to assess their automation potential.
Alba, Chief Intelligence Officer
Artificial Intelligence and Strategy Expert at Orchestra Intelligence.
Read Next
AI Agents in 2026: A Deployment Guide for France, Switzerland, and the UAE
Legal frameworks, market maturity, model selection, budget, and governance: what actually changes when a company deploys an AI agent in France, Switzerland, or the UAE in 2026.
100 AI agents per employee: NVIDIA's vision and what it changes for SMEs
Jensen Huang announces 100 AI agents per employee at NVIDIA within ten years. Karpathy confirms agents already write 80% of his code. Gartner projects 42% of companies will deploy this year. Here is what these signals change concretely for SMEs and mid-sized businesses.
5 million jobs threatened by AI: how SMEs should respond to the arrival of AI agents
The Coface and OEM study warns of 5 million jobs at risk within five years. For an SME, the real topic is not panic, it is task reorganization, upskilling, and the measured deployment of useful AI agents.