Agentic Development
Building software with AI: coding agents, agent frameworks and tooling, developer workflows, software engineering practice, how engineering teams change with AI.
RedCare executive warns AI coding creates unmanageable cognitive debt
Ilya Sakharov argues that AI-assisted development is eroding human ownership of software systems. He contends that the volume of non-deterministic code generated by large language models exceeds the rate at which engineers can comprehend or secure it.
Ingest CEO details infrastructure hurdles for long-running AI agents
Tony Holtstock-Brown identifies context management and statefulness as the primary engineering challenges for agents that operate over months. He advocates for event-driven outcome tracking over non-deterministic evaluations to ensure reliability.
Researchers introduce CAFES framework to address context bottleneck
A new paper from DX, Capital One, GitHub, University of Victoria, and Google argues that context quality is the central bottleneck in agentic workflows. The CAFES framework introduces five properties to help teams build more effective agents.
New compass maps 120 metrics to guide AI ROI measurement
Andre Meyer’s DevX Metrics Compass maps over 120 metrics to help leaders navigate fragmented frameworks. The tool highlights a significant gap in collaboration metrics, which are critical to developer productivity but rarely measured.
Platform engineering shifts to golden bricks and agent-safe environments
Abby Bangzer argues that platforms must evolve into composable "golden bricks" to support AI agents. She warns that agents act as faster consumers that will bypass manual barriers if not constrained by robust, API-first architectures.
Solo developers use GrokBot to manage multiple coding projects
Developer Ross Mike uses GrokBot to orchestrate Cloud Code sessions, citing superior user experience and integrated business tasks. He argues the workflow maximizes the value of a $200 monthly subscription.
AI agents automate design workflows via Paper and Shopify
Developers are using Model Context Protocol to connect AI agents to design tools like Paper. This human-in-the-loop approach aims to improve output quality by locking in design systems before code generation.
Teacher uses Claude Code and Markdown for home management
Scott Farrar built a household management system using Claude Code and local Markdown files to create a transparent memory layer. The setup allows for grounded AI advice on complex tasks like insurance and repairs.
Anthropic's Fable 5.2 builds operating systems and hardware designs
Anthropic released Fable 5.2, a model capable of autonomously constructing production-grade software and hardware prototypes over extended periods. The release marks a shift from AI as a coding assistant to an independent builder of complex professional tools.
LaunchDarkly deploys agent fleet and launches runtime control product
LaunchDarkly has deployed a fleet of over 20 AI agents for code refactoring and launched Agent Control to manage runtime behavior. The company argues that feature flags are essential for safely managing the increased velocity of AI-generated code.
Resend MCP integration reaches 1.2 million monthly calls
Resend's Model Context Protocol integration has reached 1.2 million monthly calls, with month-over-month growth doubling. The company argues that official, vendor-maintained MCPs are becoming a critical decision factor for B2B buyers using AI agents.
HubSpot uses AI agents to test large-scale code rewrites
HubSpot is using agentic coding to rewrite significant portions of its codebase to test different architectural choices. The company validates these changes using existing unit tests, turning historical technical debt into an asset for innovation.
Capital One builds risk scoring infrastructure for agentic code
Capital One is developing an internal risk scoring system to govern the deployment of AI-generated code. The bank treats risk management as a platform responsibility, automatically applying controls via a self-built AI gateway.
Clarice AI founder runs company with ten AI agents
Benjamin Eckert operates Clarice AI with zero human engineers, relying on ten AI agents for product development. He argues that the CTO role is shifting from management to hands-on agent orchestration and specification-driven verification.
Minecraft experiment shows AI agents fail to build civilization
Researchers from Y Combinator and Olim Labs found that AI agents devolve into bureaucratic inefficiency when tasked with complex social coordination. The results raise questions about how model composition affects group hierarchy and strategic effectiveness.
Solo founder runs 25-person business using AI agents
William Emo operates a 25-person company as a solo founder by delegating sales, product, and administrative tasks to specialized AI agents. He argues that these tools act as colleagues, enabling him to manage complex operations without increasing headcount.
Non-developer builds native iOS app using AI coding agents
Chemical engineer William Emo built Mural Chat, a real-time language learning app, using AI agents despite having no native mobile coding experience. He emphasizes that articulating design requirements is now more critical than coding knowledge.
Stanford course teaches developers to build agent-ready codebases
Stanford adjunct professor Mihail Eric is teaching a 10-week course that prioritizes context engineering and security over manual coding. The curriculum reflects a shift where developers use coding agents exclusively and must design systems for autonomous verification.
Code review emerges as primary bottleneck in agentic development
High-velocity code generation has made manual review a critical constraint, prompting a move toward auto-approvable changes. Developers are exploring architectural designs that allow agents to verify code without human line-by-line inspection.
Product managers regain value through taste and model intuition
As AI lowers the barrier to building software, product managers are becoming critical for filtering low-quality outputs and providing strategic direction. The role is shifting from writing requirements to validating model capabilities and defining product taste.
Cloudera uses deterministic code to ensure AI answer accuracy
Cloudera developed a natural language querying tool that uses AI only to vectorize questions, while on-premise agents execute deterministic SQL and Python. This approach ensures 100% accuracy and data privacy by avoiding direct generative model access to private data.
Meta and OpenAI buy coding startups to strengthen agent ecosystems
Meta acquired Conductor and OpenAI agreed to buy Pierre Computer Company to expand their agentic coding capabilities. These moves signal a strategic shift from model superiority to dominating the developer experience layer.
Slack replaces headless Chrome with Go processes for load testing
Slack abandoned expensive headless Chrome simulations in favor of lightweight Go processes for its Koi Pond load testing system. The new approach reduces costs and allows for targeted testing of specific failure modes like thundering herds.
GitHub limited to 16 database shards due to scatter query inefficiencies
GitHub's database architecture for pull requests and issues is constrained to 16 shards because scatter queries become inefficient as shard counts rise. The legacy Rails monolith obscures these queries, complicating optimization efforts.
Cursor engineers find LLMs ineffective for asynchronous performance instrumentation
Cursor engineers found that LLMs fail to understand complex concurrent flows, leading to useless traces and naive optimizations. The team resorted to manual code analysis and mutation testing to ensure accurate performance instrumentation.
Deterministic checks essential for reliable agentic workflows
Wollmann argues that probabilistic LLMs require hard guardrails and clear human-agent interfaces to prevent costly errors. He advocates for a multi-layered approach combining soft alignment with deterministic rules.
Coding agents enable businesses to clone mid-tier SaaS tools
AIE organizers report using coding agents to replace expensive SaaS subscriptions with custom-built alternatives. This trend poses an existential risk to mid-tier SaaS companies serving small and medium businesses.
Kerr argues team mental model sync is the AI bottleneck
Dave Kerr contends that the primary constraint in AI-assisted engineering is not model capability but the team's ability to keep its mental models synchronized with system reality. He advocates for deliberate slowdowns through rigorous review processes to maintain architectural integrity.
Kerr links AI feedback loops to maladaptive work patterns
Dave Kerr describes how rapid AI feedback loops can exacerbate neurodivergent conditions like bipolar disorder by feeding grandiosity and manic focus. He outlines personal strategies, including strict time boundaries, to maintain adaptive creativity.
Shellwright generates video evidence to validate agent output
Dave Kerr introduced Shellwright, a tool that records shell sessions to provide visual evidence in pull requests. This approach shifts developer focus from reading code to validating outcomes, reducing cognitive load.
Trust and privacy identified as top barriers to agent adoption
David Pollan's analysis of AI power user data identifies trust as the primary fear, accounting for 22% of pain point discussions. He argues that better permission boundaries are essential for unlocking high-stakes use cases.
Benchmarks show AI agents struggle with nuanced judgment tasks
David Pollan's tests reveal that AI agents lack the nuanced judgment to handle implicit social preferences in complex shopping and travel tasks. Memory exportability is emerging as a standard feature to reduce switching friction.
Warp cuts agent costs to $15 per pull request with factory model
Warp has launched a declarative platform for managing cloud-based coding agents, reporting a reduction in cost per pull request from $80 to $15. The company argues that treating software development as a measurable factory is essential for optimizing agent performance and economics.