Battle of the Bots: Choosing the Right AI Coding Partner

Calvin Hendryx-Parker, CTO and AWS Hero at Six Feet Up, teamed up with Travis Frisinger, Director of AI at 8th Light, for a lively “Battle of the Bots: AI Agent Showdown” presentation. The duo evaluated five AI coding agents to determine which tools can stand beside developers as true partners, and which fall short.

The session was intentionally unscripted, with live demos running against real-world projects. Instead of theory, the audience saw each tool wrestle with practical challenges, sometimes impressing, sometimes stumbling, but always offering lessons for teams weighing AI adoption.

“If you’re not already using AI coding agents, you should be,” Calvin said early on. “They’re your new pair programming partners.”

Can Your AI Agent Do This?

AI-powered assistants are becoming integral to developer workflows, but not all tools are created equal. Before relying on one, make sure it can answer critical questions:

• Can it handle complex, multi-service projects?
• Does it respect permissions and security boundaries?
• Will it save effort, or quietly drain tokens and budgets?

Which AI Agents Belong in Your Toolkit?

Calvin and Travis tested five contenders: Aider, Goose, Claude Code, Cursor, and Devin. Each reflected a different approach, from open-source flexibility to commercial polish, from semi-agentic “pair partners” to claims of full autonomy.

AI agents aren’t just a smarter autocomplete. They are autonomous systems capable of perceiving, acting, and learning across sessions.

Calvin and Travis grouped them into three categories:

• IDE-integrated (e.g., Cursor, GitHub Copilot)
• CLI-based (e.g., Aider, Goose)
• Semi- or fully-autonomous (e.g., Devin)

A recurring theme was the importance of Model Context Protocol (MCP), which enables agents to reach beyond code into documentation, browsers, and even language servers.

The results were mixed. Some agents struggled with project structure. Others impressed with orchestration, efficiency, or comprehensive test generation. One tool even spun up subagents on its own, demonstrating both how quickly the space is evolving, and how unpredictable it can be.

The verdict: there’s no single winner. Success comes from combining multiple agents, balancing autonomy with oversight, and treating AI as a sidekick, not a senior engineer.

5 Ways to Make AI Work or You

If you’re adopting AI, keep these practices front and center:

1. Experiment broadly– Different agents excel at different workflows. Many teams will benefit from running two or three in parallel.
2. Secure workflows – Trust only vetted MCP servers, enforce OAuth, and use CI/CD guardrails.
3. Guide, don’t abdicate – Treat agents as “well-educated interns.” Code reviews are non-negotiable.
4. Watch token spend – Monitor token dashboards, use tools with diff-based updates, and summarize contexts to avoid runaway costs.
5. Work in small steps – Iterative tasks are easier to guide and validate than sweeping automation.

Watch the Full Showdown

Want to see how the tools fared in real time? Watch the full presentation and review the slides.

The Battle of the Bots wasn’t about crowning a champion. It was about cutting through hype to reveal what’s real, what’s risky, and what leaders should consider as AI coding agents move from novelty to necessity.

If your team is exploring AI in software delivery, let’s talk.