Agentic Development
Building software with AI: coding agents, agent frameworks and tooling, developer workflows, software engineering practice, how engineering teams change with AI.
Chaos reports up to 10x output gains from AI agent adoption
Chaos reports that developers fully adopting AI agents produce up to 10 times more output, reducing development timelines from 18 months to 3 months. The company argues that domain expertise and industry conservatism protect its business model from AI-driven disruption.
GitHub Next develops Chopin for visual, real-time agent planning
GitHub Next is building Chopin, a tool that uses visual artifacts and real-time chat to structure plans between humans and agents. The project aims to address the limitations of text-based interfaces and improve how junior developers learn from senior engineers' agent interactions.
Architects favor self-contained systems for AI-driven full-stack development
Software architects argue that Self-Contained Systems are the optimal pattern for AI-driven development because they match the full-stack nature of coding agents. The approach offers a faster path to modernizing legacy monoliths but requires balancing team autonomy with code consistency.
Platform giants hold context edge over agentic AI developers
Agentic AI faces a strategic disadvantage against platform giants like Apple and Google due to a lack of deep user context. This limits the ability of general agents to provide proactive, personalized services compared to vertical applications.
AI agents expose legacy testing gaps and require new coordination rituals
Engineering organizations are finding that AI agents amplify existing weaknesses in testing and code review rather than creating new problems. Teams are adopting new planning rituals to coordinate multiple autonomous agents and prevent conflicts.
Figma deploys internal AI systems to automate experiments and PM tasks
Figma has built internal tools including AutoXP for growth experiments and PMOS for product managers. These systems use AI agents to automate analysis, writing, and task management, reducing human intervention in routine workflows.
AI coding agents move software delivery constraints to product decisions
Benedikt Stemmelt argues that AI agents have eliminated coding as the primary bottleneck, shifting friction to requirement clarification and organizational decision-making. He warns that without standardized processes, companies risk widening operational gaps despite faster code generation.
Developers build custom AI agent platforms for high-margin SMB niches
A developer built a custom agent platform in 36 hours using Opus 5.5, proposing a $25,000 upfront fee for small businesses. The model targets industry-specific pain points, leveraging pricing power far above generic consumer AI subscriptions.
JetBrains introduces ThinkRail agent harness for spec-driven workflows
JetBrains has released an experimental agent harness called ThinkRail that uses Git worktrees and Markdown spec graphs. The tool focuses on narrow agent workflows rather than broad general-purpose integration.
Architects must shift to enablement roles to manage AI agent autonomy
Luca Mezzalira argues that software architects must evolve into enablement teams, using domain-driven design to calibrate AI oversight. He warns that AI accelerates code generation but risks junior developer skill retention if collaborative practices are abandoned.
Arca deploys tiered models and deterministic guardrails for financial agents
Arca has built a model-agnostic agent harness that routes financial tasks to specific model tiers while enforcing strict human-in-the-loop controls. The system prevents autonomous execution by requiring manual approval for all trades and money movements.
Cosign launches agent-to-agent protocol for professional networking
Cosign is introducing an agent-native interface that allows AI agents to negotiate on behalf of users within its professional networking platform. The move aims to replace traditional cold outreach with automated, first-pass evaluations of mutual goals.
Spawn's 10-person team ships 77,000 monthly code changes via agents
Spawn, a 10-person gaming startup, shipped 77,000 code changes last month using a persistent agent fleet that costs 30 times its payroll in tokens. The company eliminated pull requests, relying on AI judges to manage decision bottlenecks and reduce human intervention.
Roblox pilots autonomous production deployments using AI risk assessment
Roblox is piloting a system that allows low-risk pull requests to merge into production without human review, using AI risk assessment and traditional engineering guardrails. The company is also redefining engineering metrics to focus on feature velocity and spec quality.
Supabase engineers report 60% of code written by AI agents
Supabase internal data shows 60% of engineers now write most of their code with AI agents, up from 10% six months ago. The shift has redirected engineering effort toward testing and re-architecting tools to be agent-friendly.
Coding agents shift developer roles to orchestration and system design
AI agents are moving from chat interfaces to execution layers capable of autonomous planning and action. This shift is redefining the developer role toward orchestration and intent definition, raising concerns about entry-level talent pipelines.
Winters urges shift from coding to design and verification in CS education
Titus Winters argues that computer science curricula must abandon manual implementation exercises in favor of high-level design and AI output validation. He advocates for deterministic tools over direct AI inference for large-scale refactoring to ensure accuracy and auditability.
Specialized agent harnesses enable non-technical founders to drive development
Bastian Buch describes a multi-agent system that allows non-technical founders to contribute to frontend development through natural language feedback. The approach uses specialized 'harnesses' to abstract complexity, avoiding the context overload associated with direct LLM prompting.
Cisco automates code reviews with multi-model agent strategies
Cisco engineers use AI agents to automate pull request reviews, with over 95% of new AI features shipped using Codex. The company employs a multi-model strategy to mitigate bias and built a unified platform in under eight months using agentic coding.
ThoughtWorks Radar highlights multi-agent orchestration and harness engineering
ThoughtWorks Technology Radar Volume 35 identifies multi-agent orchestration and harness engineering as key themes. The report advises against 'human in the loop theater' and recommends treating harness engineering as a platform discipline.
Shopify designers use AI agents to build live prototypes
Shopify's product design studio is replacing static handoffs with AI-driven development, allowing designers to generate functional code directly from Figma sketches. This shift enables small teams to execute complex visual effects and test live prototypes without traditional engineering resources.
Supabase launches compute features for long-running agent workloads
Supabase introduced Supabase Compute to run long-running workloads in the same private network as the database, removing previous time limits. The company also released pg-donta, a library that represents database state as files to facilitate agent-driven development.
Gamma CPO says agent harnesses are the new product moat
John Narona of Gamma argues that the tools and context layers surrounding large language models, not the models themselves, create value in AI products. He advises founders to focus on proprietary data and distinct vision rather than training custom models.
AI-generated pull requests strain Postgres maintainers
The Postgres core team is facing an influx of AI-generated code that is difficult to validate, creating a bottleneck for maintainers. Experts note that AI models struggle with C code, raising concerns about the quality and security of critical infrastructure projects.
Deterministic MCP servers improve AI coding agent consistency
Hello Tomorrow developed an MCP server that provides deterministic architectural constraints for AI coding agents, preventing complexity degradation. The tool parses codebases to build dependency graphs, allowing agents to verify architectural compliance before code generation.
Anthropic launches Claude Managed Agents to simplify deployment
Anthropic introduced a new API suite that allows developers to define and start agent sessions without building custom infrastructure. The platform separates execution logic from durable state to improve security and latency.
Agentic pull requests show significantly lower merge rates
Linear B data indicates that agentic pull requests are merged at a rate of 33%, compared to 88% for human-written code. The disparity is attributed to the lack of human advocacy for agent-generated code.
AI evaluations must measure customer value, not just performance
Teresa Torres argues that generic technical metrics are insufficient for assessing AI product success. Teams must define specific activation and retention metrics that reflect actual customer value.
Nois AG pivots to general market with structured AI workbench
Nois AG has launched the Cloud Solution Workbench to serve general industries, shifting its focus from high-regulated banking clients. The move replaces deterministic code generators with LLMs guided by structured models to ensure production-ready output.
Effective AI coding requires disciplined, small-context prompts
Nois AG advises that high-quality AI-assisted coding relies on breaking problems into small, atomic tasks rather than using vague, large prompts. This disciplined approach ensures consistent output and easier code review.
AI coding tools push startups toward non-technical buyers
Snyder argues that AI coding tools make DIY software viable for many users, forcing startups to target customers who cannot code themselves. This shift is critical for maintaining value against general-purpose models.
Reddit builds internal AI prototyping harness for non-engineers
Reddit has developed Reddit Proto, an internal tool allowing designers and product managers to create production-like prototypes using real data. The system uses a custom MCP to map Figma designs to production code tokens.
InnoQ defines six patterns for web frontend integration
InnoQ consultants presented a taxonomy of six patterns for web frontend integration at the Europlop conference. The framework emphasizes loose coupling and team autonomy in distributed systems.