Hiring is moving from test scores to proof of work. An Anthropic hiring manager has dropped coding interviews for his team and now looks for small prototypes in portfolios. For non-IT professionals, that opens a real door: two or three focused AI prototypes from your own field can say more than any certificate.
Two stories from the same week
Story one. On 27 October 2026, Felix Rieseberg, a hiring manager at Anthropic, wrote Hunger beats tenure: Advice for junior engineers. He said that this month he "laid the traditional coding interview to rest." His reasoning: there is "no more point in giving you a little React challenge than there is in quizzing people about assembly; Claude is now better at both."
Instead, he wrote, "I'm scanning resumes and portfolios for little prototypes that exhibit taste, empathy, and principled thinking."
An important caveat from Firstpost: this is Rieseberg's decision for his own team, not a blanket Anthropic policy. The company's careers guidance still says technical candidates may be assessed with live coding tools. Firstpost also reports that, according to an Anthropic report from 16 October, junior candidates fell from 27% of its technical hires in early 2025 to 9%.
Story two. IIT Madras's Centre for Responsible AI published a white paper, How AI Is Reconfiguring Work in India, covered by The Indian Express. Its argument: AI is not simply killing entry-level jobs, it is breaking the ladder to senior roles. Many junior roles consist of repetitive tasks that also serve as training grounds. If those are automated, people lose the way they used to build judgment.
My take: the ladder moved
Read together, these stories can sound bleak. Fewer junior hires. Fewer training tasks. Higher bars.
I read them differently. The old first rung was a credential or a test: a degree, a certificate, a timed coding challenge. Those are exactly the things AI now does well. The new first rung is proof that you can choose a worthwhile problem and build something useful, using AI.
That rung is not reserved for engineers. In fact, people from non-IT backgrounds often have the advantage that matters most: deep knowledge of a real domain and real users.
Rieseberg makes a related point about newcomers. Fresh eyes spot friction that old-timers stopped noticing. A teacher, an HR executive, an accountant or a logistics manager sees problems in their world that no engineer ever will.
Why prototypes beat certificates
A certificate says you completed a course. A prototype shows:
- Judgment. You picked a problem worth solving.
- Empathy. You understood the people who have it.
- Execution. You got something working.
- Communication. You could explain why it matters.
Those are the qualities the IIT Madras paper suggests entry-level work is shifting toward: checking AI outputs, handling escalations, evaluating how systems perform. A good prototype demonstrates all of them.
How to build your first AI prototype (non-IT edition)
This follows the spirit of Rieseberg's advice, adapted for people who are not engineers.
1. Start from a goal that matters
Rieseberg tells juniors to pick business impact over backlog. For you, that means: what does your team, school or business most want to improve? Fewer errors? Faster turnaround? Happier customers? Start there.
2. Pick one problem you see every week
Look for a task that is repetitive, annoying and well understood. Examples by field:
- Teaching: a helper that turns a rubric into personalised feedback drafts for students. (See our AI for teachers and colleges guide.)
- HR: a checklist assistant that summarises applications against a role's must-haves, with a human making the final call.
- Finance: a sheet that flags mismatches between invoices and GST records, with AI drafting the follow-up email.
- Operations: an automation that turns daily WhatsApp order messages into a clean tracker.
- Small business: a FAQ assistant trained on your own policies.
3. Keep it to one good idea
Rieseberg warns that juniors often cram "three excellent ideas among 20 bad ones" into one prototype. Do not. One problem, one clear solution. The point is to communicate an idea so well that people instantly get it.
4. Build it fast with the tools you already have
You can build a surprising amount with plain-language instructions. Some starting points:
- Chat assistants like ChatGPT, Claude or Gemini, with custom instructions or projects. Not sure which to pick? Read which AI tool should you pay for.
- Workflow automation tools that connect forms, sheets and email. Start with our explainer on workflow automation.
- Coding agents that write the code for you, if you want a small web app. See do you still need to code.
- Connectors that let AI reach your files and apps. Our Claude skills, connectors and plugins guide and what is MCP explain how.
Aim for "works for one real person" before "looks polished."
5. Use real (safe) data and real feedback
Test with a colleague or a friend in your field. Remove personal or confidential data first. Our guide on keeping your data private in ChatGPT, Claude and Gemini covers the basics.
6. Write it up in five lines
For each prototype, write:
- The problem, and who has it.
- What you built.
- One decision you made, and why.
- What it saves or improves (only claim what you actually observed).
- What you would do next.
Add a two-minute screen recording. That is your portfolio entry.
What a strong non-IT AI portfolio looks like
Two or three prototypes is enough to start. Together they should show:
- Depth in your domain. All from a field you know well.
- Range of tools. For example, one assistant, one automation, one simple app.
- Thinking, not just output. Your write-ups show judgment and trade-offs.
- Responsible use. You checked AI outputs and kept a human in the loop where it mattered.
Put them on LinkedIn, a simple personal page, or a shared folder you can link in applications.
Common objections
"I'm not technical." You do not have to be. You have to be clear about a problem and willing to iterate. Many of our learner stories started from zero.
"Recruiters still filter by degrees." Many do. A portfolio will not replace that everywhere overnight. But it gives you something concrete in every conversation, and it travels with you between jobs.
"Isn't this just for startups?" Prototypes work inside big organisations too. Rieseberg describes how two relatively new engineers built a working prototype over one weekend, and the first public version of a popular product shipped roughly eight days later. Your version might be a tool your whole department adopts.
Where to start this week
- Write down three recurring problems at work.
- Pick the smallest one.
- Spend one weekend building a rough version with an AI tool.
- Show it to one person and ask what is missing.
- Write your five-line summary.
If you want a longer plan, our 90-day AI roadmap for non-IT professionals breaks it down week by week.
Build your portfolio with guidance
At The Arthi AI Collective, I help people from every background turn real problems into real AI prototypes, with 1:1 mentorship along the way. Explore the programs, including AI From Zero and AI Career Switch.
FAQs
Do I need to know coding to build an AI prototype? No. Chat assistants, no-code automation and coding agents let you build with plain-language instructions.
Did Anthropic stop doing coding interviews? One manager, Felix Rieseberg, stopped them for his own team. It is not a company-wide policy.
What did the IIT Madras study say? That AI is weakening the ladder to senior roles by automating the routine junior tasks that used to build experience.
What should a non-IT AI portfolio include? Two or three focused prototypes from your field, each with a short write-up and demo.
How long does a first prototype take? Often a weekend, if you keep it to one idea.