Atlas survey

Live Workshop Logistics and Team Setup for an AI Coding Cage Match

Run-of-show, room and network guide, and per-tool pre-event checklist for hosting a live 4-way AI coding tool competition with real-time audience scoring.

15 sources ~5 min read #208 workshop · logistics · team-setup · ai-coding · live-event · hackathon · wifi · scoring

TL;DR Assign one tool per team of 3–5 people at least 48 h before the event — don’t let participants self-select on the day. [12] Pre-provision all subscriptions and authenticate all accounts the week before; credit-based tools (Copilot, Windsurf) can exhaust allocation mid-session. [2] Budget 5 Mbps per device with a VLAN per team, use Slido for live Q&A and voting, and run a tight 30-min build → demo → score sequence. [7] [5]

Team formation

4 teams, 3–5 people each. Pairs produce bottlenecks; groups of 4–5 allow a driver/navigator/observer split without paralysis. [6]

Assign tools pre-event. Left to self-select, participants cluster on the tool they already know — which defeats the comparison — or fight over the “best” tool. Send tool assignments with confirmation emails at least 48 h before. [13] Two strategies:

Strategy When to use
Random assignment Blind evaluation; any bias in results is chance, not skill
Skill-matched assignment Fair race; pair experienced devs with harder-to-use tools to equalise

Roles within each team:

Role Count Job
Driver 1 Uses the tool; types all code
Navigator 1–2 Reads requirements; directs the driver
Observer 1–2 Notes prompts tried, dead-ends, speed; reports to judge

Cluster teams, don’t mix rows. [1] Separate clusters prevent accidental prompt cross-contamination and make tech support routing easy.

Room and display setup

Two screens beat one. A presenter display shows the facilitator’s scoring dashboard and countdown timer; an audience display mirrors the active team’s IDE/terminal. [4]

Key settings: [4]

  • Font ≥ 18 pt in IDE and terminal; bump to 22 pt if the room is deeper than 8 m
  • Black text on white background outperforms white-on-dark when projector lumens are limited; dark themes are fine with ≥ 3,500 lm
  • Clear all notifications (OS alerts, Slack, email) on every presenting machine before the clock starts
  • Carry HDMI + USB-C adaptors for every team, plus 2 spare sets

Each team cluster needs a 4-outlet surge-protected power strip. If the building shares circuits, request a dedicated 20 A circuit per cluster from the venue.

Network infrastructure

Metric Recommended
Bandwidth per device 5 Mbps [7]
Devices per participant 1.5× headcount [8]
Users per radio ≤ 30 [8]
Network segmentation VLAN per team [15]
Standard Wi-Fi 6 / 6E preferred

For 20 participants: 30 devices × 5 Mbps × 1.4 headroom = 210 Mbps minimum on tap. [1]

VLAN per team. One team’s agentic loop shouldn’t starve the others. If the venue can’t provision VLANs, bring a managed switch (TP-Link Omada or similar) and create SSIDs per team. [15]

Failsafe. One charged mobile hotspot per team in a sealed envelope, opened only if venue WiFi fails. Brief participants upfront so there’s no panic scramble.

Per-tool pre-event checklist

Run this in the week before the event — not the morning of. [13] [1]

Claude Code [3]

  • Anthropic Claude Pro ($17–20/mo); install via npm install -g @anthropic-ai/claude-code; authenticate once per machine
  • Available as VS Code / JetBrains extension or terminal-native CLI
  • ⚠ Heavy agentic sessions can hit the 5-hour usage reset window — avoid rehearsals on the same morning

Cursor [3]

  • Download from cursor.com; Cursor Pro ($20/mo) required for Composer 2 multi-file editing
  • Supports Claude, GPT-5.5, Gemini 3.1 Pro — decide which model to demo before the event
  • ⚠ Free-tier trials expire unpredictably; use paid seats

GitHub Copilot [2] [3]

  • Pro ($10/mo) via GitHub settings, or Business ($19/seat) for org-wide rollout
  • Extension for VS Code, JetBrains, Neovim, Xcode — confirm participant IDE before the day
  • June 1, 2026: billing switched to usage-based credits (1,500/mo on Pro); agentic loops drain allocation fast — verify credit balance 24 h before

Windsurf / Devin Desktop [3]

  • Rebranded to Devin Desktop on June 2, 2026 (cognition.ai); existing Windsurf accounts carry over
  • Windsurf Pro was $20/mo → Devin Desktop Teams $40/user/mo; download new installer under Devin branding
  • ⚠ Docs still say “Windsurf” in many places — same product, new name; brief participants

Kiro [3]

  • Download from kiro.dev (AWS-backed); Pro ($20/mo) = 1,000 credits
  • Spec-driven workflow: provide a SPECS.md alongside the requirements doc to reduce cold-start confusion
  • CLI available (Windows 11+, April 2026); macOS/Linux via desktop app

Audience engagement

For a 20–100 person developer audience, Slido edges out Mentimeter on Q&A management: the upvoting module lets technical questions surface naturally. [5]

Tool Best for Cost (100 participants)
Slido Q&A upvoting, post-demo polls ~$17.50/mo, 200 participant cap
Mentimeter Live leaderboard (competition mode) €13/mo; free tier limits to 50/mo
ClassPoint PowerPoint-integrated quiz + timer ~$8/mo, up to 200

Run two audience interactions: (a) a quick vote immediately after each team’s demo (“did this tool actually solve the problem?”), and (b) a ranked-choice poll after all four demos for the overall winner. Display the join code before each interaction — don’t assume people noted it from slide 1.

Roles on the day

Role Count Key responsibility
Facilitator 1 Clock, transitions, energy
Judge / scorer 2–3 Applies rubric; resolves ties [9]
Tech support 1 per 2 teams Auth failures, IDE crashes, WiFi escalation
Demo operator 1 Runs the test suite on the projector during scoring

Judges must see the scoring rubric and calibrate against a sample output before the event starts. Publish criteria at least 24 h in advance — participants build toward what they are evaluated on. [9] For scoring infrastructure, HackerEarth integrates coding assessments directly into competition workflows; Devpost is simpler for smaller events. [14]

Run-of-show

Clock Action
T−60 min Accounts verified, repo cloned, all tests passing on each machine
T−30 min Teams seated, Slido join code on screen, WiFi per-team SSID confirmed
T−10 min Facilitator reads task spec aloud; questions answered; timer visible
0:00 BUILD starts (all 4 teams simultaneously)
+15:00 Facilitator check-in: “any tool crashed?” — surface blockers early
+30:00 BUILD stops; git commit or screenshot to lock state
+32:00 Team 1 demo: 3-min walkthrough → demo operator runs test suite on projector
+45:00 Teams 2–4 in sequence (~13 min per team)
+57:00 Judge scores finalised; Slido overall-winner poll opens
+65:00 Results announced + structured debrief

Pre-flight checklist

  • All tool subscriptions paid and active; credit/token balance confirmed 24 h before [2]
  • Repo cloned and dependencies installed on every participant machine
  • All pre-existing tests passing before the clock starts (baseline locked)
  • VLAN or dedicated SSID per team configured and tested [15]
  • HDMI + USB-C adaptors for each team; 2 spare sets for presenter display [4]
  • Slido/Mentimeter event created; join code tested from a non-presenter device
  • Scoring spreadsheet or platform pre-loaded with rubric [9] [14]
  • Printed requirements doc (1 per team) as offline fallback [1]
  • Mobile hotspot per team, charged and ready in sealed envelopes
  • Power strips (4-outlet, surge-protected) at each team cluster [1]
  • All OS/app notifications cleared on presenting machines [4]
  • Judges calibrated on sample output; scoring rubric published to participants

Citations · 15 sources

Click the Citations tab to load…