The Software Architects' Newsletter
August 2026
View in browser

Welcome to the InfoQ Software Architects' Newsletter! Each month, we bring you essential news and insights on emerging patterns and technologies from industry peers.

This month, we focus on "Building Data and AI Platforms". Technologies, patterns, and practices from this topic span the entire "diffusion of innovation" graph in last year's "InfoQ Software Architecture and Design Trends Report".

We examine how architects design data systems that balance governance with accessibility, including data mesh, decentralized ownership models, and platform engineering approaches that make data infrastructure self-service. As organizations move beyond monolithic data warehouses toward composable, product-oriented data platforms, understanding the architectural trade-offs in scalability, discoverability, and cross-team enablement is becoming a core concern.

News

Harper Argues Against the Multi-System Stack and Releases 5.2

The database platform Harper advocates for a single-runtime architecture that keeps application code and data together, with its benchmark against a Vercel-based stack reporting significantly better performance on live, personalized-data workloads. Harper recently released version 5.2, with a new record cache and more throughput per node.

Spotify Builds External Index to Enable Low Latency Point Queries on Its Data Lake

Spotify has introduced Random Access Parquet (RAP), a storage architecture that facilitates writing low-latency point queries directly against data stored in its data lake, allowing online services and AI applications to retrieve individual records without replicating datasets into operational databases. RAP adds an external indexing layer over Apache Parquet files, enabling interactive lookups while continuing to use the same datasets for analytics, machine learning, and online serving.

Netflix Open-Sources Agentic Workflow for Causal Inference

Netflix open-sourced an agentic workflow for Observational Causal Inference (OCI) that reduces toil in causal analysis. Given observational data and the human user's analysis plan, the agent uses an actor-critic loop to estimate causality, write a report, and suggest next steps.

The agent workflow builds on existing OCI tools at Netflix. The goal is to automate error-prone or repetitive tasks such as sensitivity analysis or tracking multiple iterations, leaving higher-level tasks such as framing questions and evaluating results to the human user.

The Right 300 Tokens Beat 100k Noisy Ones: The Architecture of Context Engineering

In this QCon London talk recording, Baruch Sadogursky and Patrick Debois discuss why coding agents fail due to bloated context windows and stuffed prompts. They explain practical context engineering fixes, including lazy-loaded skills, versioned context artifacts, externalized memory banks, and LLM-as-a-judge evals. Software architects & engineering leaders will learn how to turn raw markdown files into reliable agentic workflows.

How Artificial Intelligence Disrupts Engineering Progression

AI is disrupting career progression by eliminating learning opportunities at each rung while also helping people to perform above their experience level, as Alasdair Allan explained in his talk Engineering Progression When AI Ate the Middle at QCon London. Fewer junior developers are joining the industry, while AI is slowing entry-level hiring.

In a related InfoQ podcast, Michael Stiefel spoke with Tracy Bannon about the role of artificial intelligence in software and the attendant risks in security, software development, and society at large.

Case Study

An Evolutionary Architecture Pattern for Managing AI's Pace of Change

AI capabilities are evolving faster than enterprise systems can safely absorb. A recent article argues that this gap should be designed for rather than waited out. The authors, participants in the InfoQ Certified Architect Program, propose the AI gateway as an architectural seam that isolates the fastest-moving parts of AI systems, guardrails, model routing, agent identity, action policy, and audit, so the rest of the platform can remain stable.

The article explains why traditional API gateways fall short for agentic AI. Such gateways assume deterministic behavior, schema-level failures, and fully client-specified actions. Agentic systems break all three. Identical prompts can yield different reasoning paths. Failures are often semantic rather than technical. Agents interpret goals and choose their own actions rather than simply executing pre-defined requests.

The proposed AI gateway centralizes several control planes. Model routing isolates churn in provider capability and pricing. Security controls cover identity and delegated authorization, per-action zero-trust policy checks, segmentation of what agents can reach, and bidirectional content guarding against injection and data leakage. Observability and audit are treated as distinct needs; semantic logging that captures request, decision, and action, rather than plain call logs, supports both operational monitoring and regulatory evidence.

The authors are candid about trade-offs: added latency, unclear ownership across teams, the probabilistic nature of guardrails, and ongoing operational cost. They note the pattern isn't universally right. Single-team or single-model deployments may not need it.

Drawing on documented incidents, the article outlines four adoption stages organizations typically pass through, from single-team pilots to fragmented multi-provider use, then a forcing incident, and finally consolidation onto a shared gateway layer. Organizations with mature platform engineering can build by using this approach proactively. Others tend to adopt it reactively, after an incident, at higher cost.

This content is a short summary of a recent InfoQ article by Joe Price, Branimir Đurek, Pavlos Migkiros, and Trevor Dearham, "An Evolutionary Architecture Pattern for Managing AI's Pace of Change".

To get notifications when InfoQ publishes content on these topics, follow "AI, ML & Data Engineering", "Big Data", and "Database" on InfoQ.

InfoQ Online Certification Program: Time with the Practitioners You Would Otherwise Only Catch on a QCon Stage

The people who could test your thinking now report to you. This year that includes deciding when an agent's change is safe to go live, and defending that call alone.

AI-Assisted Engineering is the newest addition to InfoQ's Online Certification Program, for teams already shipping production code with coding agents. Every cohort is led by a working practitioner, someone who is still doing the job day-to-day. Five weeks, four hours a week, online with them and a small cohort of senior engineers and software architects from other companies. Each week you take a QCon talk and apply it to your own work.

Upcoming InfoQ Online Certification Programs:

• Architecture with Luca Mezzalira, Principal Solutions Architect.

  The sociotechnical side of the job: trade-offs and communication, decentralized decision-making, platform engineering, and the AI architecture decisions you're now being asked to own.

  Two intakes are open, starting September 14 and October 9. Two groups each, 09:30 BST and 10:00 EDT.

• Engineering Leadership with Michelle Brush, Engineering Director for SRE at Google.

  Moving from writing the code to setting the technical direction, and being accountable for it. Your decisions will be tested against senior peers from other companies.

  Starting September 18, Fridays 10:00 PDT.

• AI-Assisted Engineering with Zichuan Xiong, Head of AIOps, and Premanand Chandrasekaran, Head of Technology at Thoughtworks.

  Securing and governing AI systems in production: sensitive data handling, threat modeling and red teaming, controls and sandboxes, governance and auditing.

  Starting August 26, Wednesdays 2:00 PM CEST, with an October 14 intake to follow.

See all cohorts and dates.

Upcoming Events