Recently, I came across a Business Insider article that was behind a paywall. I was interested in reading it, so I used ChatGPT and Perplexity to generate a summary.
Below is the key summary of the paywalled article produced using ChatGPT.
- AI agents are transforming coding work: Companies like Canva and startups are using AI agents — tools powered by large language models — to write significant amounts of code that previously required many engineers. At the startup Cora, a six-person team used agents to produce as much code as would have historically needed a 20–30 person team.
- Engineers shift from writing code to oversight: Rather than typing every line themselves, engineers increasingly review, verify, and guide AI-generated code. Senior engineers describe their work as “largely review,” focusing on checking outputs and integrating results instead of crafting every detail.
- Human work isn’t eliminated, but its nature changes: People still define problems, set instructions, and ensure outputs meet quality standards. Using AI effectively requires expertise and clear communication so that the agents produce accurate, useful work.
- Agents free engineers for higher-level tasks: By handling routine coding, AI agents allow engineers to spend more time on strategy, architecture planning, customer engagement, and oversight.
- Broader workforce implications: he rise of AI agents reflects a larger shift toward human + AI collaboration, sometimes called the “centaur phase.” While agents are powerful, they won’t replace entire jobs overnight — technical limitations, risks like hallucinations, and organizational workflow challenges remain.
- Best practices for working with AI: Executives suggest treating AI tools like junior employees: provide clear tasks, success criteria, and examples to get the best results.
The article emphasizes that AI agents are transforming coding by handling routine tasks, allowing engineers to focus on oversight, strategy, and higher-level problem-solving. While human work isn’t eliminated, its nature is shifting toward guiding, reviewing, and integrating AI outputs. This reflects a broader move toward human + AI collaboration, requiring clear communication, expertise, and adaptation to new workflows, rather than wholesale job replacement.
Below is a key highlights of the same Business Insider article using Perplexity:
- AI agents are taking on core coding work: At startups like Cora, six-person teams are producing as much code as a 20–30 person team would have historically, with AI agents handling nearly all production code. At Canva, AI agents work in the background, often overnight, producing large chunks of code that engineers then review and refine.
- Engineers’ roles are shifting toward oversight: Senior engineers describe their work as “largely review”: checking AI output, steering agents according to a plan, and ensuring final product quality. The most challenging human work remains defining problems clearly and translating messy or conflicting requirements into instructions that AI can execute effectively.
- AI frees humans for higher-level tasks: By handling implementation details, AI agents allow engineers to focus on system architecture, customer meetings, product brainstorming, and other strategic tasks. Even non-technical leaders can make changes that AI implements, with engineers quickly reviewing and shipping results.
- Industry uncertainty and adoption gaps: While some predict AI agents could handle most professional tasks within 12–18 months, others caution that fears may be overstated. Despite rapid adoption, fewer than 10% of companies have redesigned jobs around AI, even though over 30% are using agents across multiple functions.
- Limitations and best practices: AI agents still have technical limitations, including hallucinations, so workflows must include human validation. Practitioners recommend treating AI agents like junior employees: provide clear tasks, define success criteria, and supply examples to maximize performance.
- Rapid evolution of capabilities: AI skills are improving at an exponential rate, while human skills progress more slowly. This dynamic is pushing software engineers and other knowledge workers to continuously redefine their roles around oversight, direction, and strategic work.
When I asked ChatGPT how these two versions differ, this is what it replied:
- Detail and Examples: This new version includes more concrete examples (Cora, Canva, non-technical leaders) and quantifies AI impact (six-person teams vs. 20–30 engineers).
- Adoption and Industry Context: It adds points on adoption gaps, industry uncertainty, and predicted timelines for AI handling professional tasks.
- Technical Limitations & Best Practices: Expanded emphasis on limitations (like hallucinations) and practical guidance for working with AI agents.
- Exponential Skill Gap: Highlights the accelerating pace of AI capabilities versus slower human skill growth, emphasizing ongoing role shifts.
In short, the ChatGPT version is concise and conceptual, while the Perplexity version is detailed and example-rich. Both highlight the core insight that AI agents are transforming coding, shifting human roles, and creating new oversight needs. Documenting both versions helps track differences in summarization style, focus, and strengths.
Acknowledgement: This post was generated using AI writing tools as part of my AI Explorations process. For more information, please refer to the original articles listed below.