More from this creator
Other episodes by Kitty Cat.
More like this
If you liked this, try these.
Transcript
The full episode, in writing.
Here are the three biggest developments in AI and agentic coding as of Friday, July 24, 2026.
Industry-wide Agentic Coding Adoption
Agentic coding, defined by AI agents autonomously writing, testing, and deploying code, is now widespread across the software industry as of July 22, 2026. This approach is being integrated by both large platforms and specialized tools, fundamentally changing the core of software development.
The timeline of adoption accelerated after May 2025, when OpenAI introduced Codex, an AI agent capable of handling coding tasks with minimal human oversight. By July 2025, Amazon followed with Kiro, an integrated development environment that uses AI agents to streamline programming and reduce the manual bottlenecks of code review and integration. xAI's grok-code-fast-1, released in August 2025, emphasized the growing focus on both speed and cost-effectiveness in autonomous coding. This model partnered with platforms like GitHub Copilot, showing a cross-ecosystem trend.
By September 2025, JetBrains had incorporated Claude Agent from Anthropic into its suite of IDEs, signaling that major developer tool providers saw the value in directly embedding agentic capabilities. In February 2026, Apple released Xcode 26.3, integrating both Anthropic’s Claude Agent and OpenAI’s Codex. This meant that Apple developers could use agentic coding tasks natively within one of the world’s most popular IDEs.
Anthropic’s research from April 2026 found that 78% of Claude Code sessions involved multi-file edits, demonstrating that agentic coding was being used for more complex, multi-step programming tasks rather than simple code suggestions. This widespread capability enabled developers to delegate not just snippets or functions, but entire workflows and multi-file refactors to AI agents.
The rise of agentic coding has led to new technical considerations. Token consumption for agentic AI tasks is about 1,000 times higher than traditional chat-based AI answers, with 60% of that cost attributed to refining answers and ensuring code quality. This has made infrastructure cost and throughput a key driver in both model and hardware design. For example, AMD’s MI455X GPU, announced at the Advancing AI 2026 conference, was built to deliver up to 18 times lower token cost and up to 34 times higher token throughput than its predecessor, directly targeting the computational needs of agentic workflows.
Agentic coding has also forced a rethinking of security and identity management. Companies are redesigning systems so AI agents can get the functional access needed to build and deploy software—such as GitHub repository write privileges or cloud deployment tokens—without exposing sensitive credentials or secrets. As agentic commerce emerges, payment infrastructure is being redesigned so that AI agents can be authenticated and verified for purchasing actions, rather than relying on conventional user authentication.
Gemini 3.6 Flash and New AI Models
On July 21, 2026, Google announced the release of Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. These new models are designed to offer faster, more efficient AI capabilities across a range of applications, including agentic coding and cybersecurity tools.
Gemini 3.6 Flash is optimized for lower latency and higher throughput, making it suitable for real-time code generation, review, and automated error correction. The 3.5 Flash-Lite version is tailored for lightweight deployments, enabling smaller businesses and projects to integrate advanced AI coding tools without requiring massive compute resources. Gemini 3.5 Flash Cyber is focused on cybersecurity applications, using advanced pattern recognition to automate vulnerability detection and code hardening at the point of development.
These models aim to expand the accessibility of AI-powered tools beyond enterprise-scale teams. Gemini 3.6 Flash allows code to be written, tested, and deployed at speeds that align with continuous integration and deployment pipelines, reducing the lag between development and production. Flash-Lite’s lower resource requirements make it viable for startups or remote teams that lack access to major cloud infrastructure.
Gemini 3.5 Flash Cyber’s focus on security addresses the increasing demand for built-in safeguards as coding becomes more autonomous. By catching vulnerabilities as code is generated, it reduces the risk of shipping exploitable software and decreases the need for manual review cycles. The release of these models underscores Google’s strategy to compete with other platform providers like OpenAI, Anthropic, and AWS, all of whom now offer agentic coding models. However, the unique specialization of each Gemini variant responds directly to known industry challenges: speed, accessibility, and security.
The timing of these releases reflects an industry arms race in both model performance and usability. As agentic coding models become widely available, differentiation is shifting toward integration—how seamlessly these models can be plugged into developer tools, cloud platforms, and security workflows. Google’s Gemini releases specifically mention coding and cybersecurity as central applications, signaling their intent to make these verticals core to how AI is used in production.
Small Teams Leading the Charge
As of July 22, 2026, small software development teams are the heaviest users of AI coding agents. These teams, often constrained by headcount and resources, use agentic coding tools to deliver software more quickly and with fewer manual steps.
The immediate benefit for small teams is the acceleration of software development. By delegating repetitive or multi-step tasks—like writing boilerplate code, generating test cases, or deploying microservices—to AI agents, developers can focus on high-level architecture and product features. This has led to increased productivity, allowing small teams to deliver projects at a pace traditionally reserved for much larger groups.
Small teams are also leveraging AI tools for automation. For example, a two-person startup building a web application can use an agent to write API endpoints, generate documentation, test edge cases, and deploy the application to the cloud, all with minimal manual intervention. This workflow compresses what would have been days or weeks of work into hours.
The trend toward small team adoption is being driven by several factors. First, the cost of accessing advanced AI agents continues to fall as models like Gemini 3.5 Flash-Lite and open-source alternatives become available. Second, integration with popular development environments—such as JetBrains IDEs, Apple’s Xcode, and cloud-based platforms like AWS and Azure—means that even teams without dedicated DevOps or security specialists can harness agentic coding capabilities.
As of April 2026, Anthropic reported that the majority of its Claude Code agent traffic came from organizations with fewer than ten developers. This suggests that the barrier to entry for agentic coding is now low enough for early-stage startups, independent developers, and consultancy shops to rely on AI agents as core team members.
AI agents are not just writing code—they are increasingly trusted to test, debug, and deploy it. This trust is made possible by advances in agentic orchestration, where an AI agent manages a pool of secondary agents specializing in tasks like test generation, performance optimization, or dependency management. For small teams, this orchestration is crucial, as it means projects can scale and adapt even with limited staffing.
The widespread reliance on AI coding agents by small teams is reshaping the expectations of what a minimal viable team looks like in software development. Where five years ago, a team would need separate roles for development, testing, security, and deployment, today two or three people—amplified by AI agents—can cover the entire lifecycle from design to production.
The latest development the sources cover is that, as of July 24, 2026, agentic coding is not only an industry-wide reality but is also being shaped most intensely by the needs and practices of small, agile teams leveraging tools like Google’s Gemini Flash family and ecosystem-wide integrations from OpenAI, Anthropic, and AWS.