Agentic AI: How autonomous AI agents are reshaping the workplace

From chatbot to autonomous agent

While classic chatbots mostly react to predefined dialogues, the term Agentic AI refers to a new generation of autonomous systems. These agents combine perception, goal-setting, planning, execution and learning to complete tasks on their own.

Agentic AI is much more than a conversation partner: these are AI units that act with a certain “will”. They define goals, build plans, access tools and data, and dynamically adapt to changing conditions. The difference resembles that between a calculator and an accountant: one returns a result on request, the other sees the job through from start to finish.

Three principles of the agentic mindset

For autonomous agents to work reliably, they need to master certain capabilities:

1. Initiative

Agents make their own decisions and pursue goals, rather than just reacting to inputs. When an agent is tasked with creating a report, it searches for data sources on its own instead of waiting for further instructions.

2. Adaptability

They adjust plans to unexpected events and use feedback to optimize. If one approach fails, they try alternatives, just like an experienced employee.

3. Multi-step reasoning

Complex tasks are broken into sub-tasks and worked through step by step. The agent plans ahead, prioritizes steps and coordinates different tools and data sources.

How agents work in practice: an example

Imagine giving an agent the following task:

“Analyze our sales figures for the last 12 months, compare them with the previous year and create a presentation for the management meeting.”

A classic chatbot would respond: “Please upload the data and describe which analyses you want.”

An agentic system, on the other hand:

1. Accesses the CRM and exports the relevant data
2. Cleans the data and identifies outliers
3. Performs the analysis (trends, comparisons, forecasts)
4. Creates visualizations (charts, tables)
5. Generates a PowerPoint presentation with executive summary
6. Sends an email with the finished document

The entire process runs autonomously. The human receives the finished result.

Practical examples from companies

Early adopters are already running Agentic AI in production. For teams looking for a no-code entry point, the OpenAI Agent Builder offers a low barrier to entry. The overview of AI agents compares the other platforms worth considering.

Far fewer cases are solidly documented, however, than the usual listicles suggest. Two survive a source check:

Palo Alto Networks runs an internal AI agent called “Panda” that answers employee requests to IT, HR and Finance. According to CIO Meerah Rajavel, it autonomously handles close to 60% of employee-generated tickets - a statement made to Fortune in June 2025, not an independent measurement.

Klarna announced in February 2024 that its AI assistant had handled 2.3 million customer service conversations in its first month - by the company’s own math, the work of 700 full-time agents. The story has a second half: from May 2025 onwards, CEO Sebastian Siemiatkowski publicly walked the course back, spoke of quality losses caused by the cost focus, and Klarna is hiring humans for customer service again.

Earlier versions of this article also listed percentage figures from PwC, NielsenIQ and H&M here. When we checked the sources in August 2026, none of these values could be traced back to an original source - they circulate in vendor listicles with shifting numbers and meanings, sometimes with swapped metrics. They have been removed. The pattern is itself a lesson for handling agent success figures: if you cannot trace a percentage back to its original source, do not plan with it.

The two documented cases still show both sides: agentic systems run in real business processes today - and getting there is no sure thing, as Klarna’s course correction demonstrates.

Which jobs will change?

According to the World Economic Forum (Future of Jobs Report 2025), around 78 million net new jobs will be created by 2030 - driven by all structural forces combined (technology including AI, demographics, the economy, the green transition), not by agentic AI alone. At the same time, many existing job profiles will change fundamentally.

Jobs with high automation potential

• Data entry and capture: Agents can fill forms, transfer data and validate entries
• Research and reporting: Gathering, structuring and preparing information
• Scheduling: Managing calendars, planning meetings, resolving conflicts
• Initial customer contact: Answering FAQs, categorizing requests, resolving standard cases

Jobs that will grow in importance

• AI orchestration: People who configure, monitor and optimize agents
• Quality assurance: Checking AI outputs for accuracy and relevance
• Strategic work: Creative and strategic tasks that require human judgment
• Customer relationships: Complex, emotional interactions that require empathy

The new job profile: the “Agent Operator”

A new role is emerging: the Agent Operator. This function combines technical understanding with process knowledge:

• Defines tasks and goals for AI agents
• Monitors execution and intervenes when problems arise
• Continuously optimizes prompts and workflows
• Ensures quality and compliance

What does this mean for Swiss SMEs?

For Swiss companies, specific areas of action emerge:

Short term (2025)

• Launch pilot projects: Identify a repetitive process and test an agent
• Train employees: Build prompt engineering and AI literacy
• Establish governance: Define guidelines for AI usage

Medium term (2026-2027)

• Scale successful pilots: Expand proven applications
• Redesign processes: Optimize workflows for human-agent collaboration
• Develop talent: Upskill employees into “Agent Operators”

Long term (2028+)

• Secure competitive advantage: Those who use agents effectively will be more efficient
• New business models: Agents enable services that were previously uneconomical
• Address labor shortages: Agents can fill gaps left by the skills shortage

The human role remains central

Despite all automation, the human role is indispensable. Agents need:

• Clear goals: Humans define what should be achieved
• Quality control: Humans review results and correct mistakes
• Ethical guardrails: Humans decide what may be automated
• Creativity: Humans develop new ideas and strategies

The most productive teams will be those that combine human strengths (creativity, empathy, judgment) with machine strengths (speed, endurance, consistency).

Swiss legal framework for autonomous agents

When agents make decisions that affect individual people, such as applicant pre-screening, tariff classification or automated complaint handling, Art. 21 DSG (Swiss Data Protection Act) applies: those affected have the right to a statement and to review by a natural person. For sensitive personal data (health, religion, biometric data), Art. 22 DSG additionally requires a Data Protection Impact Assessment (DPIA) before productive use.

The EDÖB (Swiss FDPIC) clearly signals in its opinions that human-in-the-loop is not just a design principle under Swiss law but a concrete obligation in workflows involving personal data. Before any productive agent deployment affecting people, the compliance review belongs in the sprint plan. More detail in the DSG guide for Swiss SMEs.

Conclusion

Agentic AI is still at the beginning, but it will rearrange the division of labor between humans and machines. The change is not a threat but an opportunity, provided companies actively shape it instead of waiting passively.

In our training programs we show you:

• How to understand the agentic mindset and put it to use
• Which processes are suitable for automation
• How to prepare your employees for the change
• How to integrate agents into your existing workflows

---

Sources:

• World Economic Forum: Future of Jobs Report (2025)
• McKinsey: The State of AI (2025)
• Klarna: AI assistant handles two-thirds of customer service chats in its first month (press release, 27 Feb 2024)
• Fortune: As Klarna flips from AI-first to hiring people again (9 May 2025)
• Fortune: Palo Alto Networks CIO on the internal agent “Panda” (18 June 2025)