AI Makes Us WAY Faster but Fries Our Focus (Here's What I Do About It)

Summary

AI coding agents have dramatically increased productivity, but running multiple agents in parallel can leave people mentally exhausted. The speaker identifies context switching as the main cause: moving between agent conversations, monitoring progress, and providing follow-up information. Because human attention does not parallelize like AI workloads, this can undermine concentration and make it harder to reach a focused “flow” state.

The speaker describes a transitional period in which agents still need frequent supervision. People may start more sessions rather than wait, increasing short-term output but also adding interruptions and pressure. Although he misses long, uninterrupted coding sessions, he says he would not return to working without AI because of the productivity gains.

Proposed Approach: Build More Autonomy, Not More Sessions

Rather than continually adding parallel agent sessions, the speaker recommends building systems that let agents handle longer tasks with less supervision. His approach includes:

• Defining the task and intent carefully with an agent before work begins.
• Codifying personal workflows, rules, and testing practices.
• Having agents plan, implement, validate, and iterate on larger tasks.
• Batching decisions and reviewing completed work periodically instead of checking in constantly.

He is building a software factory: a customized workflow in which an agent manager can break a planned release into issues and coordinate agents as they work. He argues that a tailored system can be more reliable than a simple built-in /goal feature because it can reflect how a particular person works and tests. He expects software factories and more autonomous agents to become increasingly common, while acknowledging that these systems remain limited and the future is uncertain.

The speaker also anticipates that faster LLM responses could help users stay in one primary planning and coordination session while delegating larger workflows to background agents. This might reduce the need to jump between conversations, though it remains a prediction rather than a settled outcome.

Other Observations

The speaker sometimes plays chess on Lichess or Chess.com to focus on something away from agent conversations. He also mentions wanting more deeply focused hobbies, such as reading, to counterbalance fragmented work.

Sponsor Feature: Revolut Business

The sponsored segment demonstrates using a Revolut Business virtual card for an AI agent. The card can be restricted—for example, to software purchases in the US, with a $50 weekly limit and an expiry date. These limits are enforced on the card itself rather than relying only on instructions in the agent’s prompt.

Transaction webhooks can notify the speaker’s service so it can log charges, set alerts, or take other actions. The goal is to give an agent a controlled budget without granting access to the full company account.

Guides and Tutorials Mentioned

• The speaker references another video on his channel about planning a larger software release with an agent, breaking it into issues, and using an agent manager to oversee the workflow.
• He points interested viewers to his publicly developed software factory and discusses learning how such systems are designed.

Main Speakers and Sources

• Cole Medin — Main speaker discussing AI-agent workflows and his software factory.
• Thibaut — Cited for a post about AI-driven context switching and fatigue.
• Kyle Cook — Referenced for discussing concentration and context switching.
• Cal Newport — Quoted on concentration becoming a scarce resource.
• Dynamous community members — Raised the possibility that users may simply switch between multiple software factories.
• Revolut Business — Sponsor featured in the virtual-card demonstration.