Artificial Intelligence (AI) is changing more than individual tools: it is affecting entry into the workforce, professional training, and business processes. Early-career workers, people with standardized duties, and learners who need to acquire new skills are especially exposed to the shift. Recent US data and several new products show how closely work and learning are becoming connected.
What is changing for entry-level pay
A working paper from the United States Census Bureau suggests that graduates from college majors heavily exposed to AI have started their careers on lower salaries since ChatGPT was released in late 2022. According to the summary of the findings on entry-level work and pay, starting salaries for this group were 13 percent below those of earlier cohorts. Computer science graduates are one example given in the report.
The figure describes a US development and does not prove on its own that ChatGPT caused the decline. Large technology companies also cut thousands of jobs during this period, which may have affected salaries and vacancies. The lead report further notes that the sweeping disruption once feared across the whole labor market has not yet occurred.
The signal still matters to people starting their careers. Tasks such as initial drafts, basic analysis, documentation, and standardized checks can increasingly be handled with generative AI, meaning systems that create new text or other content in response to instructions. These assignments have traditionally helped junior employees gain practical experience and demonstrate that they are ready for more demanding responsibilities.
Two representative US surveys by Epoch AI and Ipsos indicate how quickly AI is entering daily life. Between March and August 2026, the share of adults using AI on at least six out of seven days rose from 8 to 19 percent, while the share using it only one day per week fell from 17 to 10 percent. However, the survey report on daily AI use says the questions were phrased differently in the two rounds, making direct comparisons less reliable.
What new learning services can provide
As duties change more rapidly, training becomes a more regular part of working life. OpenAI is expanding its Academy with learning paths for employees, leaders, educators, students, and developers. According to the announcement of the new Academy learning paths, participants will be able to build and demonstrate practical AI skills; the available summary gives no details about prices, languages, or access in Switzerland.
Other providers are adapting educational material to current media habits. ScrollEd turns text files, dry Portable Document Format (PDF) files, and textbooks into a vertical feed containing AI-generated video, audio, text, and interactive quizzes. Swiping upward opens a new topic, swiping sideways explores the current one in greater depth, and a quiz appears at the end.
For example, a student could use the service to break a long textbook into shorter subject units before studying the details. A company could prepare a training document in a similar format, letting employees work through individual topics and test their understanding with questions. ScrollEd explicitly targets educational institutions, corporate training programs, and individual users.
Its business model combines a free consumer feed with a paid Pro subscription and annual institutional licenses. The report about ScrollEd does not provide specific prices. Short formats may make a subject easier to approach, but they do not automatically replace sustained attention to context, evidence, and conflicting arguments.
Microsoft and the University of Illinois are exploring a different approach. Their StudentSim system creates digital replicas of individual students from limited records, with the aim of simulating their typical answers and mistakes as well as their reactions to explanations from an AI tutor. These simulated learners could deliver feedback more quickly because testing tutors with large and varied groups of real learners is expensive and time-consuming, according to the researchers.
StudentSim is intended to combine two qualities: it should realistically reproduce a person’s existing ability and also revise an answer after receiving help. The description of the research project relies on claims from those involved, and it has not been independently established that the method will reliably improve real AI tutors. A simulated student is not a substitute for a real person’s experience, motivation, and feedback.
What is changing in business operations
Within companies, AI is moving from assisting with isolated tasks toward carrying out complete process steps. The accounting service Tabby imports live account data, handles client paperwork, and gives small businesses current profit-and-loss information through a dashboard. It also offers a version for accounting firms and has announced Tabby Talk, a natural-language interface that accepts ordinary written requests instead of requiring users to navigate conventional menus.
One practical example is a small business that wants to monitor its financial position without manually reconciling several accounting views. According to the provider, 5,500 small businesses were using the platform 14 months after its launch, while annual recurring revenue was around $100,000 and the team consisted of seven people. These provider figures from the profile of the Tabby accounting service have not been independently verified, and no specific customer price is given.
Supply chains may also become more automated. Multi-agent systems are groups of independently operating AI programs that coordinate separate tasks within defined boundaries. Rather than merely showing demand forecasts on a static dashboard and waiting for a human to approve every action, these systems are designed to execute selected operational steps themselves.
The title of the report on multi-agent supply-chain systems broadly says that these tools are taking over execution. Its summary is more limited, referring to targeted operational boundaries. For employees, this points more toward a shift from routine approvals to supervision, exception handling, and accountability than to the immediate disappearance of all planning work.
Pros and Cons of AI-supported work and training
Pros:
- Faster access – Short learning paths and reformatted documents can make unfamiliar subjects easier to approach.
- More current information – Services such as Tabby promise live views rather than reports updated only at fixed intervals.
- More individual support – AI tutors may be able to adapt explanations to typical errors and different levels of knowledge.
- Less routine work – Automated approvals, analysis, and documentation can leave more time for exceptions and judgment.
Cons:
- A harder start – If AI handles typical junior assignments, early-career workers may lose jobs, learning opportunities, and income.
- Unverified quality – Simulated learners and provider claims do not yet show how reliably the systems perform at scale.
- Compressed content – A feed may attract attention while reducing complex subjects to easily consumed fragments.
- Unclear accountability – As systems act more independently, people must be able to trace errors, limits, and approvals.
What this means for you
For beginners: A useful first step is to complete one clearly limited task with and without AI, then compare the results. You could ask a tool to summarize a training document, check that summary against the original, and mark any missing evidence. This teaches you not only how to operate the tool but also how a result can sound plausible while remaining incomplete.
For advanced users: You can gain more value by designing a verifiable process rather than focusing only on individual prompts. For a financial overview or supply-chain decision, that means defining the data sources, permitted actions, human approvals, and treatment of exceptional cases. Skills in professional judgment, data checking, and clear documentation become more valuable under this model, not redundant.
The mostly US-focused reports do not support equivalent conclusions about entry-level salaries in Switzerland. They also provide no reliable details about local availability, support for Switzerland’s national languages, data protection arrangements, or prices for the learning and workplace services described. These issues matter to Swiss schools and companies because a useful demonstration is not yet a dependable system for sensitive educational, employee, or financial data.
The reports do not point to one uniform future but to two developments happening together: AI reduces the effort required for some routine duties while increasing the need for training, oversight, and professional judgment. Learning services may ease the transition, but they do not solve the loss of entry-level assignments by themselves. The central open risk is whether new roles with greater responsibility will emerge quickly enough and remain accessible to people at the beginning of their careers.
Sources
- ChatGPT-Effekt: Frische Informatik-Absolventen verdienen 13 Prozent weniger als ihre Vorgänger – t3n, 2026-09-21
- Expanding OpenAI Academy with new learning paths – OpenAI, 2026-09-21
- Microsoft trainiert KI-Schüler, um KI-Tutoren besser zu machen – THE DECODER, 2026-09-20
- ScrollEd wants to turn textbooks into TikTok – TechCrunch, 2026-09-20
- Tägliche KI-Nutzung in den USA hat sich in nur sechs Monaten mehr als verdoppelt – THE DECODER, 2026-09-20
- With Tabby, a former accountant is using AI to make accountants obsolete – TechCrunch, 2026-09-21
- Multi-agent AI systems are taking over supply chain execution – AI News, 2026-09-21
Image: Matheus Bertelli via Pexels