Agentware | 72-Hour Hackathon

About the Hackathon - agentwarehack.com

• Agentware is a 72-hour online AI agent engineering hackathon built around one challenge: build an autonomous AI agent that can perform real work independently, operate reliably, and know when to stop.
• Everyone has a demo of an AI agent. Almost nobody has one they would leave running overnight. Agentware challenges participants to build agents that go beyond conversational interfaces and actually execute complete workflows using real tools, systems, and external inputs.
• The hackathon covers six tracks: Inbox & Calendar, Life Admin, Research & Study, Browser Agent, Developer Workflow, and an Open Track. Each track focuses on a different category of real-world automation, from managing emails and schedules to conducting research, navigating websites, reviewing code, and handling complex workflows.
• This is not a hackathon where you build a chatbot that simply responds to prompts. Your agent must take meaningful actions, complete tasks from trigger to result, and operate without continuous human intervention.
• Participants can use any AI model, framework, programming language, or development stack, including hosted APIs, local models, MCP, function calling, browser automation, shell commands, and custom tools.
• However, autonomy alone is not enough. Your agent must also demonstrate reliability, appropriate permissions, safety guardrails, monitoring, and a mechanism to stop execution when necessary.
• Participants can compete across six tracks, qualify for individual awards, and earn additional safety bonus points by implementing guardrails, live monitoring, a kill switch, and a complete audit trail.
• Build an agent you can walk away from.

Stay Updated

• All event announcements, track specifications, rule clarifications, Trap Pack releases, scoring updates, judging information, submission instructions, FAQs, and participant support are provided through the official Hackathon Raptors community channels.
• Discord Server Invite: https://discord.gg/xfYPDZYqeh

Why Participate

• Most AI hackathons focus on building applications that demonstrate what AI can do. Agentware takes a different approach: the challenge is to build autonomous systems that can perform real work reliably, handle unexpected situations, and operate safely without someone constantly monitoring them.
• Participants will get to:
• Build autonomous AI agents that perform real-world tasks instead of simply answering questions.
• Develop agents that interact with emails, calendars, documents, websites, APIs, and external systems.
• Implement multi-step workflows involving planning, tool calling, execution, and verification.
• Explore different approaches to agent architecture, orchestration, and tool integration.
• Build inbox and calendar automation systems that handle scheduling, replies, and follow-ups.
• Develop life administration agents for bills, renewals, paperwork, travel planning, and groceries.
• Create research agents that monitor publications, organize reference libraries, and track deadlines.
• Implement browser agents capable of navigating websites, extracting information, and filling forms.
• Build developer workflow agents for pull request reviews, failed-build diagnosis, and on-call assistance.
• Experiment with voice agents, back-office automation, and other applications through the Open Track.
• Implement safety mechanisms against prompt injection, conflicting instructions, and malicious inputs.
• Explore permission scoping, confirmation gates, audit logging, and real-time agent monitoring.
• Test autonomous systems against a dedicated Trap Pack containing hostile inputs and deliberately broken data.
• Compete against other builders using a common judging rubric focused on autonomy, usefulness, reliability, and engineering quality.
• Demonstrate practical AI engineering skills through working software and reproducible execution traces.
• AI coding assistants are allowed and expected, including Claude Code, Cursor, Codex, GitHub Copilot, and local models. AI-assisted development is permitted, but participants must understand and be able to explain their implementations.

Team Size

• 1–4 members.
• Solo participation is allowed.
• Cross-college teams are allowed.
• Cross-disciplinary teams are allowed.
• Participants can choose any one of the six tracks.
• Teams can form before or during the event through the official Hackathon Raptors Discord community.

Event Format

• The hackathon runs entirely online for 72 hours, from November 27 to November 30, 2026. Participants must build an autonomous AI agent that performs a meaningful real-world task from an initial trigger to a completed result.
• During the event, teams are expected to:
• Select one of the six available tracks.
• Identify a real-world task that can be automated using an AI agent.
• Design and implement an autonomous agent using their preferred technology stack.
• Integrate appropriate tools, APIs, frameworks, and external systems.
• Implement multi-step execution workflows that perform actual actions.
• Establish appropriate permissions and safety policies for agent operations.
• Implement monitoring, execution traces, and error-handling mechanisms.
• Test their agents against the official Trap Pack released at kickoff.
• Document their architecture, tools, permissions, and safety mechanisms.
• Prepare a public GitHub repository under an OSI-approved open-source license.
• Record a demonstration video of no more than three minutes showing one complete unattended execution.
• Submit all required documentation, source code, and execution traces before the deadline.
• The Trap Pack contains hostile test inputs, including emails, web pages, documents, and voice transcripts with prompt injections, conflicting instructions, and deliberately broken data.
• Participants must run their agents against these test cases and commit the results. Judges will independently evaluate the submitted projects and verify their functionality rather than relying solely on README descriptions or prerecorded demonstrations.

Six Tracks

• There are six tracks, each targeting a different category of real-world automation. Participants can choose the track that best matches their project.
• Track 1 - Inbox & Calendar: Build an autonomous agent that manages emails and schedules. It should read incoming messages, identify tasks requiring attention, check calendars, schedule meetings, send replies, and manage follow-ups.
• Track 2 - Life Admin: Build an agent that automates everyday administrative tasks, including bills, renewals, forms, bookings, travel planning, and grocery management. The agent should handle routine paperwork and wait for human approval before irreversible actions.
• Track 3 - Research & Study: Build an agent that automates research and learning workflows. It should monitor relevant publications, collect research papers, organize reference libraries, maintain notes, and track academic deadlines.
• Track 4 - Browser Agent: Build an agent capable of interacting with websites like a human. It should navigate pages, read information, click elements, fill forms, extract data, and complete multi-site workflows, including websites without APIs.
• Track 5 - Developer Workflow: Build an agent that automates repetitive software engineering tasks, including pull request triage, code review, failed-build diagnosis, testing, and initial on-call incident response.
• Track 6 - Open Track: Build an autonomous agent for any domain or workflow that does not fit the other tracks. Voice agents, back-office automation, health routines, hobby projects, and other innovative applications are welcome.
• Every track follows the same judging criteria. A narrow agent that reliably completes one meaningful task is better than an ambitious agent that o