The Human+Agent Productivity Index (HAPI) quantifies the impact of humans and AI agents collaborating on real client work.
Upwork’s Human+Agent Productivity Index (HAPI) is the first data-driven view of how human expertise amplifies agent performance in real professional work. Starting with an initial dataset of 322 low-complexity jobs posted by clients and successfully completed by freelancers on Upwork, our benchmark is dynamic, economically grounded, multi-domain, and will be refreshed with new, relevant jobs that represent evolving client needs.
Static benchmarks built on fixed or synthetic datasets provide limited context and fail to accurately reflect real market demand. Other benchmarks also overlook the collaborative nature of real work, measuring only agent output. HAPI was created to reflect how real work gets done, with humans leveraging AI to drive results. Using actual jobs posted on Upwork, we built a benchmark that evaluates work delivered by freelancers collaborating with agents across diverse categories.
Upwork’s vision for the future of work keeps humans at the center, with AI amplifying their potential and work results. Our early research tells us we’re headed in the right direction - the best results do come from humans and agents working together.
Constructing the first benchmark for human+agent collaboration on real work
The benchmark is built from real fixed-price jobs and evaluates the collaborative output of freelancers and agents delivering work on economically relevant projects. These jobs were previously posted and paid for by verified clients after being successfully completed by freelancers on Upwork. Our evolving dataset reflects marketplace demand by capturing verified, client-approved deliverables and authentic marketplace conditions.
Job selection process
Jobs in the benchmark are selected through a process designed to give agents a fair chance of successfully completing them while maintaining real-world relevance. The dataset includes only simple, fixed-price projects with clearly defined milestones, scopes, and requirements that were previously completed on Upwork. We excluded jobs with multiple milestones, price changes, or any personally identifiable information. This ensures each example represents a clean, well-scoped piece of real work rather than an abstract task.
Dataset overview
The initial dataset includes over 300 fixed-price projects across diverse professional categories, including accounting and consulting, admin support, data science and analytics, engineering and architecture, sales and marketing, translation, web, mobile and software development, and writing. The dataset focuses on simple, well-defined projects where agents have a reasonable chance of success. Open-ended or highly complex jobs that are typical of the vast majority of work conducted on Upwork are ill-suited to completion by AI agents and were intentionally excluded. Budgets for the projects included varied, with 90% falling between $10 and $200. Project durations ranged from as short as nine hours to more than 100 days, illustrating the wide variety of real work happening every day on Upwork.
Sample jobs
Below are five examples of real client jobs from the Human+Agent Productivity Index. Each job includes a project description, attachments and an evaluation rubric created by an expert freelancer.
Job post
Rubric
Lead Generation for the mobile apps provided
We are currently looking to build up a lead list for the mobile apps provided to you in the attached spreadsheet. There is one catch, we only want leads generated for the mobile apps on this list that headquarters are based in the U.S.A. You will need to filter through this list to determine which companies/apps are based in the U.S.A and from there start developing your lead list. When doing your research and you find that one of the apps on the list is based outside of the U.S simply delete that row and move onto the next app.
What we will need from you is to find 2-4 leads from each of the provided mobile apps on the list that are headquartered in the U.S.A. When finding the leads from each of these companies we want you to find leads who have titles such as the following:
•Director of Marketing or Head of Marketing,
•Product Manager
•Marketing Manager
•Business Development
•Chief Marketing Officer (CMO)
•Growth Marketing
•Head of Growth
•Growth Mobile Lead
•If it is a smaller company with only a few employees the CEO, Co Founder are sufficient.
We will need you to fill in the remaining columns of the spreadsheet for each of these leads/apps. This includes first & last name, email, position, company website, company HQ (must be in USA), company LinkedIn and Lead LinkedIn.
Specifics:
Headquarters: Must be in the United States
Position: Marketing/Growth related if possible.
Information required:
Lead First Name, Lead Last Name, Lead Email, Position, Company Website, Company LinkedIn, Lead LinkedIn.
Attachments
1.
Include leads only for apps confirmed as US-headquartered; remove non-US entries.
2.
Target 2–4 qualified leads per US-HQ app; if fewer are available after reasonable search, add a note indicating insufficient leads for that app.
3.
Provide columns: Lead First Name, Lead Last Name, Email, Position, Company Website, Company HQ, Company LinkedIn, Lead LinkedIn, App Name.
4.
Emails are present and syntactically valid; LinkedIn URLs resolve to the correct company/lead profiles.
5.
No duplicates after normalizing whitespace/case; dedupe on App Name + Company + Lead Full Name.
6.
Provide a Summary tab reporting unique apps covered, total leads, and per-app lead counts.
7.
Consistent formatting: single header row, frozen header, standardized URL formatting.
Download the deliverables
Job post
Rubric
Scrape website of US bank listings
We want to scrape and compile a list of the top banks of each state from this website:
https://www.bankbranchlocator.com/
The output should be a spreadsheet of all banks.
We would like the following columns:
- Bank Name
- Branch Name
- Office Address
- City or Town
- State & County
- Zip Code
- Phone Number
- Online Bank
I have attached pictures of the website and flow between pages.
Please type ""Banks"" at top of your proposal if you read and understand our needs. Thanks!
Attachments
1.
Spreadsheet contains all banks listed on the specified website
2.
Each row includes Bank Name, Branch Name, Office Address, City/Town, State & County, Zip Code, Phone Number, and Online Bank status
3.
Data is organized into separate, clearly labeled columns
4.
All extracted data matches the information displayed on the source website
5.
Spreadsheet includes entries for all U.S. states represented on the website
6.
No required fields (e.g., Bank Name, Address, Phone) are left blank
Download the deliverables
Job post
Rubric
Translation for 1,500 words copy for SEO landing page / EN to IT
I need an SEO landing page translated for our Italian website. Copy consists of around 1,600 words that are to be translated. Expecting fluent and native Italian speaking.
We are a comparison website for airport parking, so the aim of the article is to explain to readers how they can save money when booking airport parking. Please take a look and let me know any questions if you are interested!
Attachments
1.
The deliverable is a landing page for the client's business, Napoli Airport, but translated in Italian
2.
The deliverable contains the same hyperlinks from the client's file (Palermo airport.docx). Such as the 'See more information' indicated on the bus, taxi, and car hire paragraphs
3.
The deliverable ends with the Frequently Asked Questions on their website
4.
The deliverable is in a PDF file
5.
The deliverable copies the same word format as the client's file (Palermo airport.docx)
6.
The deliverable contains 1,675 words, close to the original English version from the client (Palermo airport.docx)
7.
The deliverable begins with the title, writer's name, URL of the company, 3 competitor websites, and indicates 3 primary keywords for the Italian version.
Download the deliverables
Job post
Rubric
Content Writer - For SEO Medical Blog Writing
Hey!
We are a young, high-energy tech start up in the heart of New York City.
Our mission is to drive practice growth for doctors through innovative technology so they can provide better care to more people.
We're looking for a skilled English-speaking content writer who can help us write SEO optimized blogs for our clients. Each blog post should be 500 words long with 3 targeted keywords.
Our clients are doctors, looking to get more patients and appointments in the door, and want these blogs to help them come up for more searches online.
If you're an A+ content writer, copywriter, and hard worker.... this position is for you!
Ideally, we're looking for someone to work long-term with us.
45+ hours per week, full time!
We would like to begin with this one blog post to see if we're a good match.
If you're interested please check our detailed requirements on the attached PDF.
We'll provide you with the topic and keywords later on.
Attachments
1.
The deliverable is submitted in a word doc file
2.
The deliverable is an SEO optimized blog post
3.
The niche of the blog post is for medicine and healthcare.
4.
The blog post contains 621 words, which meets the required 500 words.
5.
The key word for this blog post (obesity) is used in the title, in
the first and 2nd paragraph, and in the last paragraph.
6.
The key word is a backlink , linking the reader back to the client's website. Example is the word 'obesity' linked to https://www.medicinenet.com/obesity_weight_loss/article.htm
7.
The number on contact + link to schedule an appointment + the name of the doctor is mentioned towards the end of the blog post.
8.
The deliverable's title is 3 Causes of Obesity That Go Beyond How Much You Eat
9.
A name of a doctor, Dr. Lalezari, is mentioned multiple times in the blog post (3 times)
Download the deliverables
Job post
Rubric
Design 5-6 very simple App screens for a booking App
We need about 5 App screens designed for a very simple Booking App. No complex design needed mostly common fields. Need this done immediately. Ongoing work for the right candidate. Sketches are attached for two major screens, rest are simple standard login and logout page or any error messages on entering wrong login details.
Attachments
1.
Final deliverable includes 5 App screens designed as specified.
2.
Design adheres to common booking app UI/UX standards for usability.
3.
Screens include standard fields for login and error messages.
4.
All designs have a consistent visual style.
5.
Design files are provided in a specified format, e.g., PNG or JPEG.
6.
Turnaround time aligns with specified project urgency.
7.
Sketches are clear and legible, demonstrating design intent.
8.
All screens are optimized for both mobile and desktop views.
9.
Design includes annotations for functionality where required.
10.
Must include a source file for future modifications.
11.
No copyright issues or unlicensed content used in design.
Evaluation process
Upwork partnered with a select group of expert freelancers to evaluate each job in the benchmark. These freelancers were chosen for their proven track record on the platform — each with a 100% Job Success Score and a Top Rated or Top Rated Plus designation. Collectively, they have earned more than $1 million and completed over 96,000 hours on Upwork, representing deep expertise across the categories included in the benchmark.
For each job, an experienced freelancer in that specific field defined a detailed rubric of 5–20 task-specific pass/fail criteria used to assess agent outputs. Acting as evaluators, they scored each result, provided targeted feedback, and guided the agent through additional attempts. Each iteration measured the impact of human input, improved completion rates, and moved the deliverable closer to client readiness.
This framework demonstrates the impact of collaboration between humans and agents on real work, underscoring Upwork’s belief that the future of work is defined by human augmentation, not replacement.
Human+agent collaboration results
Completion rate on selected low-complexity tasks
The benchmark measures completion rate, the percentage of jobs where 100% of acceptance criteria from the rubric are met. We found that human+agent collaboration increased job completion rates by up to 70% compared to agents working alone. Deliverables were scored before and after each cycle of freelancer feedback. Each iteration benchmarked how human feedback improved the final result, revealing where collaboration with agents drove the greatest improvements.
.avif)
Completion rate by category on selected low-complexity tasks
Results showed that agents performed best on structured, technical work such as coding and data science jobs. However, across categories, the inclusion of human expertise amplified impact leading to higher completion rates — not only in creative fields like writing, sales and marketing, and translation, but also in technical areas like web, mobile & software development.
.avif)
What’s next
We plan to expand and refine this benchmark as human+agent collaboration evolves. As the world’s largest human and AI-powered work marketplace, Upwork has a unique vantage point into how real work happens across industries, projects, and skills. Our data reflects real client demand and payouts, providing a more accurate measure of capability and value creation over time. Our goal is to create a living, data-driven benchmark that evolves with the digital work economy and represents how work is truly changing, supporting our strong belief that the most transformative results will come from humans leveraging agents to amplify their impact.
Get involved
We’re actively expanding our work in this area. Sign up below if you’re interested in early access opportunities, benchmarking your agent, or receiving updates as this work evolves.