Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tendem: A Hybrid AI+Human Platform

Konstantin Chernyshev, Ekaterina Artemova, Viacheslav Zhukov, Maksim Nerush, Mariia Fedorova, Iryna Repik, Olga Shapovalova, Aleksey Sukhorosov, Vladimir Dobrovolskii, Natalia Mikhailova, Sergei Tilga

TL;DR

Tendem addresses the challenge of producing reliable, high-quality knowledge-work outputs efficiently by combining an autonomous AI agent for routine tasks with Human Experts for uncertain or high-stakes steps. The system uses a multi-layer, gated workflow with continuous automated and human QA to prevent errors and omissions. In a 94-task in-house benchmark, Tendem achieves 74.5% Good results and cuts median delivery time by 53% relative to a freelance baseline, while lowering median price to 32 USD. External benchmarks show the underlying AI agent is competitive on web browsing and hard knowledge tasks, supporting the viability of a hybrid AI-human platform for professional remote-work tasks and providing a path for reproducibility through released benchmarks.

Abstract

Tendem is a hybrid system where AI handles structured, repeatable work and Human Experts step in when the models fail or to verify results. Each result undergoes a comprehensive quality review before delivery to the Client. To assess Tendem's performance, we conducted a series of in-house evaluations on 94 real-world tasks, comparing it with AI-only agents and human-only workflows carried out by Upwork freelancers. The results show that Tendem consistently delivers higher-quality outputs with faster turnaround times. At the same time, its operational costs remain comparable to human-only execution. On third-party agentic benchmarks, Tendem's AI Agent (operating autonomously, without human involvement) performs near state-of-the-art on web browsing and tool-use tasks while demonstrating strong results in frontier domain knowledge and reasoning.

Tendem: A Hybrid AI+Human Platform

TL;DR

Tendem addresses the challenge of producing reliable, high-quality knowledge-work outputs efficiently by combining an autonomous AI agent for routine tasks with Human Experts for uncertain or high-stakes steps. The system uses a multi-layer, gated workflow with continuous automated and human QA to prevent errors and omissions. In a 94-task in-house benchmark, Tendem achieves 74.5% Good results and cuts median delivery time by 53% relative to a freelance baseline, while lowering median price to 32 USD. External benchmarks show the underlying AI agent is competitive on web browsing and hard knowledge tasks, supporting the viability of a hybrid AI-human platform for professional remote-work tasks and providing a path for reproducibility through released benchmarks.

Abstract

Tendem is a hybrid system where AI handles structured, repeatable work and Human Experts step in when the models fail or to verify results. Each result undergoes a comprehensive quality review before delivery to the Client. To assess Tendem's performance, we conducted a series of in-house evaluations on 94 real-world tasks, comparing it with AI-only agents and human-only workflows carried out by Upwork freelancers. The results show that Tendem consistently delivers higher-quality outputs with faster turnaround times. At the same time, its operational costs remain comparable to human-only execution. On third-party agentic benchmarks, Tendem's AI Agent (operating autonomously, without human involvement) performs near state-of-the-art on web browsing and tool-use tasks while demonstrating strong results in frontier domain knowledge and reasoning.
Paper Structure (17 sections, 2 figures, 4 tables)

This paper contains 17 sections, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. System Overview
  3. Roles
  4. End-to-End Workflow
  5. Target Advantages
  6. Evaluation
  7. In-house benchmark with real-world tasks
  8. Systems Evaluation.
  9. Analysis of Quality Criteria.
  10. Price and Speed.
  11. Takeaways.
  12. External Benchmarks (Tendem's underlying AI agent)
  13. Evaluation limitations
  14. Safety, Intended Use, and Limitations
  15. Primary risks and mitigations.
  16. ...and 2 more sections

Figures (2)

  • Figure 1: The fraction of tasks rated Bad/Mediocre/Good for each system on in-house benchmark. Each system evaluated by a human without system name indication. Decline label means safety filters of the system declined the task.
  • Figure 2: Evaluated system quality vs median Total Time. Light-orange corner highlights the desirable region (lower time, higher quality).