Can Wabi write your CV? What the free AI assistant does

What exactly is Wabi, and why are job seekers trying it on their CV?

Wabi is a free personal AI assistant from Replika founder Eugenia Kuyda that opened to everyone on 19 October 2026, and it organises your life into separate chat threads that can build small custom apps on demand. That thread structure is the whole pitch. Instead of one endless conversation where context piles up and answers get vaguer, you keep a health thread, a family thread, a work thread — and the agent decides what tooling each one needs. Connect your email and it spins up an opening set of threads and apps for you. It's downloadable on iOS, Android and Mac, free for now, with a $20M pre-seed behind it and no settled business model yet.

The backstory is unusually long. Twelve years ago Kuyda stood on the TechCrunch Disrupt stage and pitched IO, a conversational concierge meant to book your restaurants and holidays — years before the deep-learning work that made today's assistants possible. That startup, Luka, became Replika, the companion app she ran until stepping down as chief executive last year. Wabi started life as something narrower: a social platform for prompt-made mini apps, pitched as the "YouTube of apps", where you'd build a tool and share it with friends. The 2.0 release flipped that around. People liked having personal software, Kuyda found, but didn't want to spend their evenings building it.

Job seekers reach for a new assistant before they reach for anything else, and a CV is the first file they feed it. The pattern is predictable: paste the job ad, paste your CV, ask for a rewrite. That instinct isn't wrong — drafting is genuinely what these tools do best — but it quietly skips the step that decides whether you get an interview, which is whether a machine can read the file you upload. If you've already tried handing an entire job search to an AI agent, you'll recognise the gap between a confident chat answer and a document that survives screening.

Can Wabi actually write your CV bullets and cover letter?

Yes. Wabi drafts CV bullets, professional summaries and cover-letter text about as well as any current general-purpose assistant, because that's a plain language task running on the same class of model underneath. Ask it to turn "managed the onboarding process" into something a hiring manager stops on, and you'll get three or four usable options. Ask it to cut a 900-word cover letter to 250 while keeping your two strongest proof points, and it will. The ceiling isn't the model's prose — it's the raw material you hand it. Vague input produces the smooth, metric-free writing recruiters have trained themselves to skim straight past.

Where Wabi does something a single chat window doesn't is the thread. Set up one job-search thread and your work history, target titles, salary floor and the roles you've already applied for live there permanently. You stop re-pasting the same three paragraphs of context into a fresh conversation every evening, and the assistant stops forgetting that you left teaching in 2023 or that you're only considering remote work. Because the agent can build a small app inside that thread, you can keep a live application tracker — company, role, date sent, stage, follow-up — sitting next to the drafting. Across a four-month search and 60 applications, that's a genuine time saving.

The writing still needs you on top of it, and this is where most drafts come apart. An assistant with no access to your performance reviews will reach for plausible-sounding numbers: "improved efficiency by 30%", "led a team of 12". If you didn't do that, you've just put something on your CV you can't defend in a first interview — and in sales, finance and healthcare, verification is routine. It also flattens the specifics that make you hireable: the exact SIEM you operated, the migration you owned, the cohort size you taught. Keep a one-page file of real figures and paste it into the thread before you ask for a single bullet.

What can't Wabi do for your CV?

Three things Wabi can't do: score your CV against anything, tell you whether an applicant tracking system will read it, or hand you a formatted document. It's a conversation, not a screening tool. There's no corpus of market CVs to rank yours against, no parser running on the file you'd actually upload, and no document export. So if you ask for a number out of 100 or an "ATS match percentage", it will cheerfully produce one — and that number is invented. This is the single most important thing to understand before you trust any general assistant with a live application.

The parsing problem is a file problem, and no chat assistant can see it. A PDF doesn't store lines or paragraphs; it stores fragments of text with coordinates on a page. Extraction has to rebuild reading order from geometry, and in a two-column template that reconstruction breaks — your skills column gets stitched into the middle of a job description, and the system concludes your employment history is incomplete. Workday-style pipelines pull text sequentially across the page; others merge adjacent blocks because the PDF carries no reading-order tags. Wabi only ever sees the text you typed into chat, never the scrambled version a recruiter's system receives. A free CV check on the actual file is the only way to see it.

Then there's the document itself. Wabi gives you text in a thread, or a small app interface — not a CV file with consistent margins, a parse-safe heading order and an export your target market expects. Copying chat text into Word and nudging the spacing by hand is exactly how good content ends up in a layout that breaks. If you want the writing and the file in one pass, build the CV in something that exports it properly: six templates, PDF and Word, and Hebrew CVs written in Hebrew and exported right-to-left — which matters if you're applying in Israel and your file has to clear a bilingual screen first.

Which prompts get the best CV output from Wabi?

The best results come from one dedicated thread, primed once with your full history and the exact text of the job ad, then asked for a single deliverable at a time. Open with: "This thread is my job search. Here's my complete work history with dates, team sizes and the numbers I can actually prove. Remember it. Never add a figure I haven't given you." Then, per application: "Here's the job ad text. List the ten skills and tools it names, in its own words, and tell me which are missing from my history." Context first, narrow request second — that split beats any single mega-prompt every time.

From there, work bullet by bullet. "Rewrite this bullet so it leads with the action, names the tool, and ends with a measurable outcome. Use only the numbers above. Give me three versions, each under 25 words." For a gap: "Write two factual sentences explaining a 14-month career break for childcare. No apology, no spin." For a pivot: "Which of my teaching responsibilities map to project management? Name the overlap, don't sell it." And since Wabi can build an app inside the thread, ask for an application tracker with company, role, date sent, stage and follow-up date — the one thing a plain chatbot won't keep for you.

Three requests waste a message. "Make my CV better" — with no job ad and no constraints, you get adjectives. "Write my CV from scratch" — you get a template stuffed with assumptions that take longer to correct than to write yourself. And "score my CV out of 100" or "what's my ATS match rate?" — fabricated, every single time, because nothing inside the assistant is measuring anything. The same ceiling applies to every agent currently being pitched at job seekers; what these agents really do when they tailor a CV is worth reading before you delegate a live application to one.

What checks should you run on anything Wabi writes?

Run four checks on anything an assistant writes: every number is one you can defend, the file parses cleanly, the job ad's real keywords appear in your own words, and the whole thing still sounds like you. Do them in that order, because the first two are what actually cost you interviews. A bullet with a borrowed metric collapses in the room. A beautifully written CV in a two-column layout collapses before a human ever opens it. Checks three and four are about surviving a recruiter's first skim, which lasts seconds and treats generic phrasing as noise.

The voice check matters more than people expect. Assistant drafts share tells: "spearheaded", "cross-functional synergies", three consecutive bullets opening with the same verb, and a summary that could belong to anyone in your industry. Recruiters clock the pattern even when no software flags it, and the phrases and formatting that give an AI draft away are quick to fix once you can see them. Worth saying plainly: invisible watermarking in chat output isn't what exposes a CV — the mechanics of why a watermark never reaches a recruiter's system explain that. The giveaway is almost always the writing.

Then do the part Wabi structurally can't. Upload the finished file — PDF, Word, or even a photo, with Hebrew OCR if your CV is in Hebrew — and you get an overall score out of 100 rated Weak through Excellent, five category scores covering experience, tech stack, impact and ownership, clarity and structure, and ATS compatibility, plus a visual layout analysis of how the page actually reads. That's free. Paste the job description in and the analysis lists the skills you're missing and generates ATS-friendly rewritten CVs in six templates mirroring the ad's keywords. The full fix list, likely rejection reasons and a hiring-probability estimate cost $3.99.

Frequently asked questions