we need to talk..

Summary

The video examines how AI agents are changing software work and hiring, using Anthropic as its main example. The narrator says an Anthropic hiring manager has stopped using traditional coding interviews on their team, arguing that coding exercises and algorithm questions are less informative now that Claude can perform them well. The broader takeaway is that employers may increasingly assess candidates on their judgment, problem-solving, and ability to deliver business value—not just their ability to write code.

Anthropic’s reported changes

• Fewer junior technical hires: The video cites figures showing junior technical employees’ share of technical hiring falling from 27% to 9%. The narrator says the company appears to favor experienced engineers, while noting that “junior” can include people with several years of experience.
• More output per employee: The video cites an average of $17 million in annual revenue per employee and claims agent work is being produced at roughly six agent-hours per human work-hour.
• Managing agents is becoming a job in itself: Employees must decide which agent tasks need attention first, monitor results, and handle urgent work. The narrator compares this to watching a stove while less urgent tasks finish on their own.
• More coordination and review: The video describes an increase in “meta-work” around Claude—from roughly 10% to 20–22% of sessions—and a rise in agent instructions from 300 to 8,500. It also says agents increasingly check other agents’ work.
• AI use across departments: Examples include sales teams processing more leads, administrative tasks such as classifying and routing messages, and AI-assisted work in engineering, security, recruiting, and other functions. The narrator says administrative roles are not growing much.
• Reported productivity gains: The video cites a 5.1× increase in commits after December 2025, along with other claimed productivity increases across teams. These are presented as figures from the material being discussed, not independently verified in the video.

Advice for early-career developers

The narrator’s central advice is to connect engineering work directly to business goals:

1. Understand what outcome matters. Identify the problem a manager or business needs solved, especially one that could increase revenue or improve a meaningful metric.
2. Build a focused prototype quickly. The hiring-manager advice discussed in the video favors ambitious ideas that can be tested over a weekend, rather than spending months on a large project.
3. Show initiative and a fresh perspective. The video recounts a Notion example in which a junior developer pushed for an offline mode despite the architectural difficulty. The narrator argues that newcomers may spot problems experienced insiders overlook.
4. Make decisions, not just code. As AI generates more possible designs and implementations, developers still need to choose the right approach for the user and project.

AI prompting and task-setting guide

The video recommends giving AI agents concrete, bounded tasks:

• Use SMART goals: make requests specific, measurable, achievable, relevant, and time-bound. For example, ask for a defined latency improvement by a stated deadline rather than a vague request to “make the code better.”
• Break work into tasks that can be assessed on a realistic timescale, and use sub-agents for bounded checks or stress tests.
• Ask for concise, structured outputs—for example, a one-line description of each sub-agent’s task and goal. The narrator refers to a prompting format as “ASDSTE 100”; the subtitle spelling may be inaccurate.

Granola product demonstration

The narrator also demonstrates Granola, an AI note-taking app, describing it as a “second brain” for capturing and organizing information. Features shown or described include:

• Voice transcription and meeting-audio transcription
• Turning brief notes into structured notes using an “Improve notes” action
• Filling in context around headings from spoken or written input
• Connecting Granola with Claude through MCP so Claude can use the stored context

This segment is a product demonstration and promotion, rather than an independent product review.

Main speakers and sources

• Main speaker: The Singh in USA narrator, who presents the analysis and advice.
• Referenced source: An Anthropic hiring manager’s blog post about ending traditional technical interviews on their team.
• Other material cited or discussed: Anthropic hiring and productivity figures, a Notion offline-mode anecdote, and a prompting approach attributed to Andriy Karpathy.