From The New Stack · 36 stories
Pilot Protocol Launches Platform and Agent App Store to Enable Inter-Agent Communication
Pilot Protocol has launched its platform, which includes an agent App Store, to facilitate communication and interaction between software agents. The platform provides agents with network addresses, allowing them to discover and utilize other agents, tools, and applications to enhance their capabilities. This development aims to move beyond isolated agents to foster an "agent economy" where agents can autonomously interact.
P99 CONF Talks Detail Database Trade-offs in Storage Design and Performance
Five talks from P99 CONF 2026 discussed database storage design, focusing on the trade-offs between read amplification, write amplification, and space amplification. These presentations covered how different database systems and engines, such as LSM trees and ScyllaDB's Logstor, manage these factors to optimize for specific workloads and storage types like object storage.
AWS Developer Advocate Proposes Enhancements for MCP Server-Agent Interactions
James Ward, a principal developer advocate at AWS, suggests that MCP servers should provide agents with guidance on subsequent actions rather than just data. This approach aims to improve agent efficiency and enable better tracing of agent interactions, moving MCP beyond a basic API to a more integrated interface.
AI Code Generation Tools May Devalue Software Features
The widespread adoption of AI code generation tools like Claude Code, Cursor, and Copilot is leading to a reevaluation of software development strategies. When code output becomes free and abundant, the value of individual features may diminish, prompting organizations to reconsider their investment in these tools.
Low Trust in Synthetic Data's Referential Integrity Hinders Enterprise Adoption
A recent Perforce Delphix study indicates that only 34% of enterprise leaders trust synthetic data's referential integrity, and 38% question its realism. This lack of trust in data quality is identified as the primary barrier to wider adoption, despite synthetic data's potential for protecting sensitive information and accelerating development.
Pi integrates MCP with Codemode to reduce token usage in AI agents
The Pi coding agent now integrates the MCP protocol, addressing previous concerns about high token consumption. This integration uses Codemode to dynamically load tools, significantly reducing the initial context window size required for browser automation tasks.
Adrian Cockcroft Discusses AI's Impact on Performance Engineering
Adrian Cockcroft, a veteran in system architecture and optimization, discussed how AI is influencing performance engineering during a P99 CONF interview. He highlighted the shift from deep kernel analysis to new data analysis methods driven by AI, emphasizing the need to look beyond standard tool outputs. This perspective offers insights into evolving approaches for identifying and resolving performance bottlenecks in complex systems.
Buildpacks standardize container image creation for improved enterprise security at scale
Buildpacks offer a standardized approach to container image creation, which helps enterprises enforce security controls more effectively across diverse development environments. This standardization addresses common issues like inconsistent Dockerfiles and uneven update cadences that hinder vulnerability management in large organizations.
Report: Developers exhibit addictive behaviors with AI coding tools, managers exacerbate issue
A new report indicates that 80% of developers view their relationship with AI coding tools as a dependence, with 43% delaying breaks and meals due to tool use. Managers are contributing to this by rewarding high AI usage, potentially leading to developers shipping code they do not fully understand.
DeepSeek V4 Flash Vision Exp vs. Gemini 3.7 Flash: A Comparison of Vision Models
DeepSeek released V4 Flash Vision Exp, its first vision model, which is compared against Google's Gemini 3.7 Flash for image input capabilities. The comparison focuses on performance, cost, and speed for tasks like chart reading, invoice auditing, and incident diagnosis. This analysis helps developers choose between two budget-friendly vision models based on specific application needs.
Anthropic's Files API saves time but not money compared to pasting for LLM prompts
Anthropic released its Files API and browser-use toolset, allowing developers to upload documents once and reference them by ID in subsequent requests to large language models. A test comparing this method to pasting content directly into each prompt found that while the Files API saves development time, it does not reduce token usage or cost.
Observability faces data storage challenges exacerbated by AI-generated telemetry
The observability industry struggles with cost-effectively storing and analyzing full-fidelity telemetry data, leading to incomplete system visibility. The rise of AI systems generating more logs, traces, and metrics is expected to worsen this existing data problem, increasing costs and blind spots for teams.
AI Agent Traces Transition from Telemetry to Application Data
The article discusses how the execution records of AI agents, initially considered telemetry, are becoming application data due to requirements for durability, retrieval, and display in product workflows. This shift impacts storage decisions and how agent behavior is managed and audited.
Anthropic's New Playground Tool Compared to OpenAI's Incumbent
Anthropic replaced its Workbench prompt-testing tool with Playground, a stateless version that removes features like saved prompts and version history. This change aligns with OpenAI's decision to shut down its own saved Prompts and Evals platform, indicating a shared industry conclusion that prompts should reside in code rather than web consoles. The article compares the functionality of Anthropic's new Playground with OpenAI's long-standing Playground.
Challenges in achieving real-time AI at scale due to data pipeline issues
Real-time AI systems often face performance degradation in production due to data pipeline problems, not model inaccuracies. Issues like tail latency spikes under concurrent load and stale feature data significantly impact accuracy and system responsiveness. Addressing these challenges requires architectural considerations rather than simple fixes.
Mistral's Shift to MCP for Data Retrieval Requires Manual Migration and Raises Data Governance Questions
Mistral is transitioning its Vibe Work platform from an internal indexed data system to a new MCP (common interface) based retrieval method for Google Drive and SharePoint, requiring administrators to manually install replacements. This change moves data retrieval off-platform, raising concerns about data governance and how third-party connectors handle customer data and permissions.
DNS Management Should Adopt Infrastructure as Code Principles
DNS management should be treated with the same operational rigor as other infrastructure components, such as servers and cloud resources. This approach, termed "domain management as code," aims to prevent outages caused by misconfigurations and improve accountability.
Cloud Security Teams Struggle with Alert Overload, Not Negligence, Says IOmergent Founder
Jon Rose, founder of IOmergent, states that cloud security teams are overwhelmed by alerts and lack the operational capacity to act on them, rather than ignoring security. This issue persists despite increased adoption of Cloud Security Posture Management (CSPM) tools, which generate numerous findings but do not solve the problem of prioritization and remediation.
Personalization is a Ranking Problem Requiring Architectural Solutions
Effective personalization in products is fundamentally a ranking problem, not a quality issue, and often fails due to architectural limitations rather than signal collection. Systems need to weigh user intent, item quality, user history, availability, and business priorities simultaneously and in real-time to deliver relevant results. The challenge lies in combining these diverse signals efficiently during query-time ranking.
Pollen Robotics and Hugging Face launch Microduck, an open-source, retrainable robot
Pollen Robotics, in collaboration with Hugging Face, has launched Microduck, a small, open-source robot designed for reinforcement learning. Users can retrain its behaviors in simulation and deploy them on the physical robot, with all software components available under an Apache-2.0 license.
AI Coding Tools Lead to Decreased Code Quality and Increased Incidents
The adoption of AI coding tools has led to a decline in pull-request review quality, an increase in incidents, and more bugs per developer, according to a report from Faros AI. This suggests that the push for "software factories" using AI without human oversight may be premature and detrimental to codebases.
Becoming an AI Company: Four Non-Technical Leadership Lessons
This article outlines four leadership lessons for companies transitioning into AI-driven organizations, emphasizing that the most challenging aspects are not technological. The lessons focus on aligning AI initiatives with business objectives, simplifying processes before applying AI, redesigning workflows, and measuring relevant results. The author argues that AI serves as an operating model to achieve strategic goals, rather than being a strategy itself.
AI Security Requires an Offensive Mindset, Not Just Defensive Measures
The Five Eyes intelligence alliance warned that frontier AI models will transform cyber capabilities, making traditional defensive security approaches insufficient. Security professionals need to adopt an offensive mindset, using AI to proactively identify vulnerabilities before attackers exploit them.
Developer perspective on avoiding vendor lock-in through open-source approaches
This article discusses how vendor lock-in often starts with reasonable choices but can lead to significant limitations and increased costs when business conditions change. It emphasizes that the real risk lies in unexamined dependencies that become too expensive or impractical to unwind, hindering a business's ability to evolve.
Code Review Evolves Beyond Bug Catching to Focus on Collaboration and Judgment
Code review is shifting from primarily bug detection to a process focused on collaboration, knowledge sharing, and verifying product suitability, especially as AI generates more code. This change redefines its role in software development, emphasizing judgment over line-by-line inspection.
Hyperscaler AI Infrastructure Spending: Growth, Profitability, and Efficiency
The article analyzes financial data from major cloud providers (AWS, Google Cloud, Azure) to assess the financial implications of their AI infrastructure spending. It concludes that growth is accelerating, profitability scales with size, and capital expenditure efficiency is improving, countering criticisms about excessive spending.
The Rise of Platform Engineering in Software Development
The role of software engineers is shifting from writing code to building tools that enable machines to write code, leading to a convergence of product and platform engineering. This change is driven by the increasing use of AI in code generation, which necessitates more focus on shared toolchains and workflows to prevent fragmentation.
Automating Code Verification to Accelerate Software Delivery
The increasing speed of AI-generated code necessitates a shift from human-centric code review to automated verification processes. This change allows development teams to ship code faster by reducing reliance on manual checks, moving humans from directly reviewing every change to overseeing automated systems.
Graph RAG Improves Evidence-Based Answers by Incorporating Relationships
Graph RAG is a method for Retrieval Augmented Generation (RAG) that incorporates relationships between facts as evidence, addressing limitations of vector search alone. This approach is beneficial when the connections between data points are crucial for answering complex queries, such as identifying service dependencies or customer contract requirements.
Strategy for Reassigning Resource Ownership Before Personnel Changes
This article provides a strategy to identify and reassign infrastructure resource ownership when personnel change roles or leave, preventing stale ownership records. It outlines a method using database queries to flag inactive owners, ensuring resource accountability and security. This approach addresses a common issue where infrastructure ownership updates are often overlooked during personnel transitions.
Attaching Owners to Cloud Resources for Improved Governance
This article outlines a method for identifying and assigning owners to cloud resources to prevent orphaned instances and reduce unnecessary costs. It proposes using continuous inventory syncing, policies to block untagged resources, and audit trails to maintain clear ownership.
Implementing Response Caching Reduces LLM Costs by Reusing Prior Answers
Response caching, an established technique from data pipelines, can significantly lower costs for Large Language Model (LLM) applications. By fingerprinting LLM inputs and dependencies, developers can reuse previous model responses when inputs have not changed, avoiding redundant token charges.
Strategies to manage tracing data overload for system failure detection
Tracing data provides insights into system failures but can lead to data overload due to storage costs and difficulty in analysis. Techniques like head sampling, tail sampling, and dynamic sampling can mitigate these issues. These methods help optimize data collection and retention, making tracing more practical for identifying system problems.
Five Steps for Building Service Architecture and Operational Resilience
This article outlines five steps for designing service architecture to improve operational resilience and incident recovery. It emphasizes the importance of clear service mapping for both human and AI-driven incident response systems. The guidance helps organizations minimize service disruptions and accelerate root cause analysis.
Routing Keys Isolate Kafka Consumer Tests on Shared Brokers
Testing Kafka consumers in shared environments is challenging because traditional request routing methods do not apply to asynchronous message queues. A pattern using routing keys, producer stamping, and consumer filtering can isolate changes for testing without duplicating the entire stack. This approach allows developers to validate consumer changes against realistic systems efficiently.
Webinar to discuss scaling Postgres for AI applications with TimescaleDB
A webinar on October 21 will feature Matty Stratton from Tiger Data discussing how to extend PostgreSQL with TimescaleDB. The session will address challenges faced when scaling vanilla Postgres for high-volume data and AI applications.