InfoWorld publishes six-environment tour of AI coding IDEs and agents

What happened

InfoWorld published a feature article titled "A brief guide to AI-powered software development environments" on Aug 10, 2026, and the publication is credited to Martin Heller in a "Contributing Writer" byline. The piece carries an "18 mins" read tag in its header, signaling a survey-style walkthrough rather than a short news item. InfoWorld positions the article under its development-environments coverage, framing the topic as "toolkits for directing coding models and coordinating agents."

Readers can find it on the Developer Tools section, alongside the broader Developer Tools Insights category where related agent-IDE developments continue to surface. Inside the feature, Heller states he will "provide a brief tour of six of them: GitHub Copilot, Google Antigravity, JetBrains Air, Kiro, Zed, and Zenflow," establishing that the article's scope is bounded to those six products rather than the broader market of "perhaps a dozen" IDEs and "another dozen" agent harnesses Heller mentions. Because the article is a publisher-authored roundup, not a vendor announcement, individual product details should be read as the report's framing of each tool.

The feature is the frozen Event; any product roadmap implication attributed to InfoWorld itself would overstate what the article claims.

Confirmed facts about GitHub Copilot

The article's most concrete, version- and date-anchored claims concern GitHub Copilot's history, modes, and billing change. Heller reports that GitHub Copilot was "First offered as a plug-in to Visual Studio Code in 2021" and today runs in Visual Studio, Vim, Neovim, the JetBrains IDE suite, the GitHub Copilot CLI, the GitHub CLI extension, and on the GitHub website itself. He lists "three default operating modes: Plan, Ask, and Agent," describing Plan as multi-step research and outlining, Ask as read-only chat, and Agent as one that "edits files in your workspace."

On tooling, the article says Copilot has "nine categories of default tools used by agents," with additional tools coming from connected Model Context Protocol servers and plug-ins; Heller adds the report-specific observation that "My installation currently has 346 tools available," alongside "16" custom modes on his machine. The single highest-impact operational change Heller restates is the billing model: "Effective June 1, 2026, GitHub tracks your token consumption rather than your Premium Request Units," with the carve-out that "Code completions and next edit suggestions are exempt from token limits."

Treat each of these as InfoWorld's report of vendor behavior, not as InfoWorld's own policy statement.

Reader impact for working developers

For developers already paying for Copilot, the headline operational consequence is the June 1, 2026 shift from Premium Request Units to tracked token consumption, with completions and next-edit suggestions explicitly carved out as exempt. That changes how heavy "Agent" mode sessions, which Heller notes "can run for hours with only a few requests for permission," will be billed relative to quick inline completions. The feature also surfaces a practical risk: a single Copilot installation can now aggregate 346 tools across nine default categories plus MCP servers and plug-ins, which raises the cost of reviewing what an agent can actually touch before granting long-running permissions.

Readers choosing between the six surveyed environments — Copilot, Google Antigravity, JetBrains Air, Kiro, Zed, and Zenflow — should weigh Heller's framing that the field has consolidated around agent harnesses that "compile and test the improved code, iterate on that, and then come up with something better." Developers prototyping Copilot CLI authentication against GitHub Apps in the same window can sanity-check tokens with a local helper such as Generate a JWT for a GitHub App with before they wire agents into billing-metered sessions.

Uncertainty and what to watch

The article is a publisher-authored feature, so product-specific claims should be read as InfoWorld's characterization rather than vendor-issued release notes. Several details in the lead of the supplied text are truncated mid-sentence — for example, the Google Antigravity paragraph ends at "The original Google Antigravity (from" — so feature-level claims about Antigravity, JetBrains Air, Kiro, Zed, and Zenflow are not grounded in the supplied evidence and are omitted from this briefing on purpose. The Copilot tool count of 346 and the 16 custom modes are explicitly Heller's "my installation" observations, not a product specification, and will vary by plug-in set.

The June 1, 2026 token-based billing change is reported as effective, but the article does not document grandfathering, plan tiers, or overage mechanics. Watch for GitHub's own release notes confirming the Premium Request Unit retirement, for vendor documentation on how long-running Agent sessions are metered, and for MCP server defaults that determine whether third-party tools count toward token consumption. Security posture around agent permissions remains an open question, especially in light of adjacent reporting on trojanized developer repositories at Netskope discloses trojanized GitHub repositories pushing AI-themed SmartLoader that agents could be steered toward.

Bottom line

The frozen Event is a feature article, not a vendor release, and the briefing above treats it that way throughout. InfoWorld's Aug 10, 2026, 18-minute piece by Martin Heller is a tour of six AI coding environments — GitHub Copilot, Google Antigravity, JetBrains Air, Kiro, Zed, and Zenflow — and the only dated, version-anchored claims it supports are Copilot's 2021 VS Code debut, its three default modes (Plan, Ask, Agent), the nine-category default tool surface with 346 installed tools in Heller's setup, and the June 1, 2026 switch from Premium Request Units to token-based billing with completion and next-edit exemption.

No claim in this briefing infers a number, a date, a deprecation, or a vulnerability that the article does not state. Numbers used here are drawn only from the frozen figure allowlist and from direct quotation of the cited Claims. Product-specific details beyond Copilot are not asserted because the supplied evidence pack truncates those sections.