When a hiring manager opens 80 portfolios on a Tuesday morning, most look the same. AI-assisted scaffolding, the same shadcn template, three "AI-powered todo app" projects, and a README that opens with "built with Next.js, TypeScript, Tailwind, and Vercel." The work isn't bad. It's just indistinguishable.
That's the filter you have to clear in 2026. Not "can you build something" — anyone with a Cursor subscription and a free weekend can. The filter is whether a reader can tell what you contributed, what you decided, and whether the thing works for anyone other than you.
What AI noise actually looks like in a portfolio review
Watch how a hiring manager scans a portfolio. They skim the homepage. They click one project. They check three things in this order: live URL, the README's first paragraph, and the commit history.
The reasons a portfolio gets closed in under 30 seconds:
- The live URL 404s or loads a Vercel preview-deploy splash.
- Every project README starts with "This project is a [type] built with [stack]" — a generated opener.
- Commit history is one squashed "Initial commit" followed by three "Update README.md" entries.
- Three projects on the homepage, all CRUD apps with "AI-powered" in the title.
None of these are deal-breakers in isolation. Stacked, they read as "I prompted a model to scaffold this and never came back."
The signals that survive the filter
Portfolios that get a second look share four traits. They're not flashier. They're more specific.
A live thing used by someone who isn't you. One project with 14 real users beats five projects with zero. The "someone else" doesn't need to be paying customers — a Discord bot your D&D group uses, an internal tool your bootcamp cohort adopted, a Chrome extension with 30 installs. The bar is external validation of usefulness, not vanity metrics.
Decision artifacts, not just code. A short writeup of why you chose SQLite over Postgres for this app, why you abandoned the first architecture after week two, what you measured before adding caching. Three paragraphs, not a thesis. Reviewers can tell within a section whether you actually made the decision or borrowed someone else's framing.
Specific, small numbers. "Reduced first-paint by 340ms after replacing the Markdown parser" reads stronger than "10x faster." Real performance work produces specific milliseconds, specific bundle-size deltas, specific p99 numbers. Round-number claims read as fabrications now.
A maintained main project. One repo with commits this month, an issues tab someone has touched, dependencies that aren't a year out of date. The signal isn't "I built this once" — it's "I'm still the kind of person who maintains things after the dopamine fades."
Most working engineers use AI assistants for portfolio projects in 2026 — that's not the problem. The trap is shipping the artifact without showing your fingerprints. If your README, commit messages, and code style all match Cursor's defaults exactly, the reader can't tell whether you understand what you shipped. A CHANGELOG, a decisions doc, or commit messages that explain why — not just what — fix this in an afternoon.
A structure that holds up
Here's a portfolio layout that consistently survives the 2026 filter. Three projects, in this shape:
One flagship. A single project that's substantially yours, has been live for at least three months, and has measurable usage you can talk about. Front-load this on the homepage. Show the live link first. The case-study writeup matters more than the design of the site.
One artifact of curiosity. Something weird — a CLI tool, a custom static site generator, a thing you built because you wanted to understand a concept. This is where you demonstrate that you read documentation for fun, ship things without permission, and have taste. It doesn't need users. It needs a clear "why I built this" paragraph.
One collaboration. A contribution to an open-source project, a hackathon team project (with a clear note on what you specifically built), or a freelance deliverable you can point to. This proves you can work in a codebase you didn't author. Without it, the rest of the portfolio reads as "smart hobbyist."
Three projects, three signals. Anything beyond that is dilution.
What to cut
Some of this used to be common advice. None of it ages well into 2026.
- The "skills wall." Logos of React, Node, Postgres, AWS, Docker, Kubernetes, Redis. Everyone has the same wall. It's signal-free. Replace it with two sentences about what you actually reach for and why.
- The "About" page with a stock photo and three sentences about coffee. Either write something specific or remove the page entirely. Generic about-pages are negative signal.
- Course completion certificates. Reviewers do not weight these in 2026. Replace each one with a link to what you built after finishing the course.
- Blog posts that are LLM-summarized tutorials. A reader can tell within a paragraph. Either write from real experience or remove the blog tab.
- "AI-powered" as a product descriptor. It tells the reader nothing. Describe what the product does for the user instead.
Open your portfolio in an incognito window and start a 90-second timer. Can you find: a working live URL, the date you last committed to your flagship project, and one specific number describing how the project performs or who uses it? If any of the three is missing, fix that before anything else on the site.
A portfolio in 2026 isn't a gallery — it's an audit trail. The work has to demonstrate that you make decisions, ship things, maintain them, and hold specific opinions backed by specific experience. That's the part AI can't generate for you, and it's the part reviewers are now trained to look for first.
Originally published at pickuma.com. Subscribe to the RSS or follow @pickuma.bsky.social for new reviews.
Top comments (0)