Anthropic engineer says AI agents could reshape office work across America

Anthropic’s Boris Cherny sounds an alarm

Ever feel like your inbox, docs, and chats run your day? Boris Cherny at Anthropic says AI “agents” will soon run many of those clicks for you, and that could shake up office life fast. His warning reflects a broader shift, with agent tools moving from demos into real-world workflows.

Cherny helped build Claude Code, Anthropic’s coding-focused agent tool that can read code, edit files, and run commands. Anthropic also introduced Claude Opus 4.6 in early February 2026 and said it was designed to handle longer agent-style tasks more reliably.

Claude Code shows what agents can do

Think of a chatbot as a talker and an agent as a doer. Boris Cherny, who leads Claude Code at Anthropic, says the jump from text replies to real actions is a significant change. That is why this change feels larger than a normal product upgrade.

Anthropic’s broader agent push points toward tools that can handle multi-step office tasks, while Claude Code is aimed at coding workflows such as editing files, running commands, and reviewing code. Anthropic says Opus 4.6 is better at planning and sustaining longer agent-style work.

Anthropic warns the shift may feel painful

Work changes slowly, then all at once. On Lenny’s Podcast, Boris Cherny, the creator and head of Claude Code at Anthropic, said AI agents could spread across most computer-based work and that the transition may feel painful for many workers.

The larger point is that agent tools may change how many computer-based jobs are done, not just how software is written.

His tip is to try the tools rather than avoid them. Start with low-risk tasks like summaries, drafts, and checklists. Learn what it does well and where you still need to double-check before you send anything to a teammate.

Why “software engineer” may fade in 2026

Job titles feel permanent until they are not. Cherny has suggested the label “software engineer” may start to fade in 2026 as more people can build software with agent tools. That does not mean coding stops; it means who does it changes.

When agents handle testing and boilerplate, teams may value product judgment more than syntax. You could see titles like “product builder,” “workflow designer,” or “automation lead” pop up. If you write requirements, spot bugs, and ship safely, you will still be needed, even if the job name shifts.

Project managers get an agent sidekick

If you manage projects, agents could become your fastest helper daily. Instead of chasing updates, you might ask an agent to scan tickets, summarize risks, and draft next steps. That can free you to focus on the human parts, like decision-making and coaching.

But it also raises the bar on clarity. Agents follow instructions, so fuzzy goals create messy outputs. Strong prompts look like good briefs: context, constraints, examples, and a definition of done. The best managers may become “systems thinkers” who design workflows that people and agents share without confusion.

Design and marketing get faster research

Design work is not just about pretty screens; it is about problem-solving. Agents can speed up research by collecting notes, comparing competitors, and turning feedback into themes. They can also generate quick copy options and A/B test ideas for ads.

The risk is bland sameness. If everyone uses the same tools, everything can start to sound alike. Great designers and marketers will stand out by bringing authentic taste, story, and audience empathy. Use agents for the grunt work, then spend your time on the choices that make people feel something.

Hiring changes when agents read resumes

Hiring could change faster than most people expect. Agents can sort resumes, schedule interviews, and draft job posts in minutes. That helps small teams move quickly, but it also can bake in unfair filters if nobody checks the logic.

Innovative HR teams will keep humans in the loop. They may use agents to surface patterns, then review decisions with clear rules and audits. For job seekers, the lesson is simple: make your resume easy for machines to read. Use clean formatting, strong keywords, and specific measurable outcomes, not vague fluff.

Support teams use agents as copilots

Customer support is part bot, part human, and agents push it further. An agent can read a ticket, pull account history, suggest a fix, and draft a reply in brand voice. That can cut wait times and reduce questions.

Still, people want empathy when they are stuck. The best support teams will use agents as copilots, not replacements. They will let agents handle tracking numbers and policy links, while humans handle exceptions and challenging emotions. If you work in support, build skills in troubleshooting and de-escalation, because that stays valuable.

Security rules tighten for computer agents

Agents that can click and run commands are powerful, and that cuts both ways. On the upside, they can review logs, spot patterns, and file alerts faster than a human. On the downside, a setup mistake could leak data or trigger unsafe actions.

Companies will tighten rules around what agents can access. Expect permission gates, audit trails, and sandbox spaces for safe agent work. For workers, the habit to build is simple: never paste secrets into tools, and review what an agent is about to send or change.

Productivity jumps, but quality matters

If agents do the busywork, productivity metrics will get weird. A single person may ship what used to take a small team, and managers may think headcount can shrink overnight. That is a risk, because speed without quality can lead to costly mistakes.

Cherny has said his team uses AI daily to move faster, and he expects improvements to keep coming. The better question is not “how fast,” but “how safe and correct.” Teams will need stronger review, testing, and clear ownership. If an agent writes it, a human still signs it.

Little-known fact: McKinsey’s 2025 research says AI is more likely to reshape how work is done than simply eliminate jobs, with workers shifting time toward judgment, coordination, and interpreting results.

Entry-level training needs a new plan

Entry-level jobs are where people learn the basics by doing simple tasks. If agents take those tasks, companies must rethink how beginners get trained. Otherwise, you get a gap: fewer juniors ready to become seniors.

One fix is guided practice with agents. New hires can run an agent, then explain the output, verify facts, and correct mistakes, like a driving student with an instructor. Another fix is rotating juniors onto customer problems, not just paperwork. If you are starting, keep a portfolio that shows judgment, not just speed.

Build agent literacy without coding skills

You do not need to be a programmer to use agents. List three annoying tasks you do each week, like formatting slides or writing status notes. Then test an agent on a low-stakes task to see quick wins.

Treat it like a new coworker. Give context, ask it to show steps, and require a final checklist before you accept results. Keep sensitive data out of prompts, and save what you asked and what it answered. Over time, you build agent literacy, which makes you harder to replace.

Want another labor market angle on how AI and automation may affect hiring? Dive into recent job-cut trends and what is driving them.

The office of the future is human plus agent

The shift may be how teams divide work. Humans will set goals, handle ethical dilemmas, and make calls when trade-offs get messy. Agents will execute, document, and repeat processes without fatigue.

Cherny says society needs to talk through the impact because the tools are moving fast. Expect workplace rules, such as what an agent can access, how outputs are reviewed, and who is accountable.

If you’re watching how media jobs are shifting, the related piece on the Washington Post’s deep layoffs adds essential context.

What do you think about Anthropic’s warning that AI agents could reshape white-collar work across America? Share your thoughts in the comments.

This slideshow was made with AI assistance and human editing.

Read More From This Brand:

• Musk says the U.S. is certain to go bankrupt without AI
• U.S. businesses spend more on equipment as AI growth offsets weakness in housing
• Feds single out Colorado in first attack on state AI laws