At AWS Summit, agents are making software pros use their words

With AI advancements, software professionals may need to brush up on natural language more than programming ones.

For Feroz Sheikh, chief information and digital officer at agricultural-tech company Syngenta, this supposedly new focus on using natural language to program is a callback to a concept called “literate programming,” introduced more than four decades ago by Stanford researcher Donald Knuth.

With a “literate” development approach, you explain the program in natural language, like an essayist, using both words and code snippets. This forces programmers to concentrate not on instructing a computer, but explaining to humans what a computer can do.

“My programs are not only explained better than ever before; they also are better programs, because the new methodology encourages me to do a better job,” Knuth wrote in a 1984 paper.

Like shoulder pads and power suits, Sheikh sees this ’80s approach making a comeback, as AI easily accomplishes the what, while the human developer plays the essential role of considering the why. “But exactly why you wrote [a particular code loop] is often lost in that deterministic instruction to the machine,” he said.

Sheikh and other tech leaders spoke with IT Brew at Amazon’s AWS Summit, hosted in New York City on June 17, where the company announced new agentic tools, features, and form factors. Those automations, aimed at making life easier for developers, leave software pros with important tasks involving the communication of intent.

Proud to announce. Here are a few notable tools announced by AWS this week:

• AWS Continuum. The tool, available in gated preview, finds code vulnerabilities, prioritizes them, and offers remediation support.
• Kiro, available on iOS. This prompt-to-spec development tool is now offered right from your iPhone.
• Amazon Bedrock Managed Knowledge Base. With this feature (and an agentic “retriever” of relevant context) developers build GenAI apps using their own data sources.

Agents at work. So, what’s left for a developer to do when AI finds your bugs, builds your app, and connects you to your company’s relevant data repos?

It’s not uncommon for David Yanacek, a senior principal engineer at AWS working on agentic AI, to have a half-dozen coding agents running at once—one building a feature, maybe others doing a fix or carrying out other tasks. For fine-tuning his plain-language or command-line specifications, he relies on Kiro, which turns specifications (descriptions of desired requirements, designs, and tasks) into a kind of blueprint for the resulting code generation and testing.

With much of the work automated, Yanacek finds himself spending more time considering how an agentic solution should behave, and reviewing specifications to get to that solution: “I’m doing a lot of figuring out what to do and how I want to go about it.”

In an October 2025 survey from software-security company Sonar, which polled over 1,100 global developers, the tasks ranked most highly effective by the developers included writing documentation, explaining or understanding existing code, and vibe-coding or creating new projects with mostly AI-generated code. Three out of four developers said that AI reduced their time spent on “toil work.”

Asking around. At the AWS event, Cristina Pieretti, general manager and head of digital content and innovation at data, analytics, and intelligence firm Moody’s, said she sees developers freed up by AI to handle important challenges like examining outputs and preventing hallucinations.

“Maybe you’re going to do less coding, for example, but you have to make sure that you have the right governance on that coding,” Pieretti said.

Jimmy Hatzell, CEO and co-founder of AI platform Hatz AI, sees IT professionals becoming more like a business-process manager, that is, a consultant focused on systems design rather than “the technical nuts and bolts.”

Hatzell looks for a more abstract skill from today’s software professionals: an imagination. “As organizations adopt AI, being able to rethink processes and how things are done in an AI‑first way becomes very important,” he told us at the AWS event.

Syngenta has deployed its share of agents for tasks like helping farmers with seed selection or sorting out supply-chain bottlenecks. Those kinds of automations have led to an emerging skill that Sheikh wants on his teams: linguistics. If an agent receives a poorly written policy document, for example, it could misinterpret the data.

Humans need to put the instructions into clear language.

“Developing the agent is not enough. You need people who are doing this work of reviewing it, identifying gaps or issues in that policy document, either rewriting it or having a conversation with the business partners that you need to change the policy,” Sheikh said.