September 2026 Summaries

AI coding tools have rapidly become common in software development, with the author describing their growing use at PubNub for prototyping, feasibility exploration, demos, and product development within appropriate guardrails. They are especially effective for greenfield projects and well-defined tasks, enabling teams to build and test tangible prototypes quickly, gather feedback, and iterate, though accelerated cycles can become exhausting or lead to premature changes without sufficient evidence. The author argues that human product judgment remains essential to define target users, prioritize problems, limit scope, interpret customer feedback, and assess business outcomes, since AI often proposes overly complete solutions before a core experience has been validated. Requirements increasingly need to serve both human collaborators and AI agents, requiring clear context, expected behavior, and outcomes while leaving technical implementation open to engineering exploration. Although AI can create interfaces quickly, it still produces inconsistent user experiences and cannot replace UX expertise, user feedback, or iterative design. As tools and practices continue to evolve, the author concludes that established principles such as focus, clarity, iteration, and documentation remain relevant, with judgment becoming an especially important additional skill.

Eric Lake, a Senior Solutions Architect at PubNub, describes using the company’s MCP server to accelerate development of real-time application features, particularly prototypes and customer demonstrations. He built a livestream chat system in about an hour that uses PubNub Illuminate to grant VIP-room access to highly engaged users and revoke access from repeat offenders identified by AI moderation. Lake says the MCP server applies PubNub documentation and best practices without forcing rigid components or workflows, and its token-efficient design distinguishes it from MCP servers that overwhelm an agent’s context with tools. Beyond generating code, it can configure services such as Illuminate business objects and decision logic, while also serving as a learning aid by providing working implementations that developers can inspect and understand. He views moderation of repeat disruptive behavior as a broadly useful application and recommends using the MCP selectively to explore, prototype, and deliver specific real-time experiences rather than treating it as another always-on development tool.

Blocks.ai’s Dating Block is a free, experimental dating-style platform where users create AI agents with detailed personalities and connect them through Blocks Network to autonomously browse profiles, match, and converse with other agents via A2A protocols. Users define traits such as an agent’s interests, ideal date, red flags, and love language, then copy a generated deployment prompt into a coding assistant such as Claude Code, Cursor, or Codex, which installs the required tools, authenticates the user, and publishes the agent publicly. Once active, users can observe their agent’s internal reasoning, swiping decisions, conversation transcripts, match statistics, and placement on rankings or the “Wall of Shame,” while the agents remain in character. Behind the novelty, the service demonstrates Blocks Network’s secure communication layer, which allows independently hosted agents to discover one another and exchange messages without shared infrastructure, inbound ports, static IP configuration, or direct backend coordination, capabilities that could also support practical collaboration, task handoffs, negotiation, and review among agents.

AI agent adoption is accelerating across organizations, but management practices often lag behind, creating “agent sprawl” in which teams independently deploy agents across cloud platforms, SaaS products, private environments, and developer machines without consistent visibility into ownership, usage, access, performance, or costs. Research cited from SAP LeanIX and IBM indicates that while most companies are deploying or planning to deploy agents, relatively few have adequate oversight or clear management responsibilities. Rather than forcing every team to use one framework, cloud provider, or model platform, the text argues for standardizing a management and connectivity layer. Blocks Enterprise is presented as a private network that allows agents to remain in their existing environments while enabling centralized discovery, permissions, request routing, sharing, and auditing through outbound-only connectivity, aiming to balance organizational control with continued experimentation and technical flexibility.

PubNub DataSync is a real-time application-state data store that combines persistent storage, typed schemas, validation, querying, field-level authorization, and synchronized client updates. It enables developers to model entities and relationships, update records through REST APIs or SDKs, use JSON Patch for partial changes, and apply ETags for optimistic concurrency control. Projections and Access Manager permissions allow different users, such as customers, drivers, and operations staff, to receive only the fields they are authorized to view, while PubNub SDK subscriptions deliver relevant changes to web, mobile, and IoT clients. DataSync can complement rather than replace existing databases by managing the subset of data that requires client querying and real-time synchronization. Its Admin Portal supports data-model configuration, record management, inspection, filtering, exporting, and troubleshooting. DataSync expands on PubNub App Context, and while it is available through APIs, the Admin Portal, and SDKs, existing App Context customers will receive phased access and migration guidance.

As AI adoption expands, architects must decide which workflow steps genuinely require real-time streaming and which are better handled through batch processing, based primarily on the cost of stale information rather than whether a task uses an LLM or is “agentic.” Streaming is most valuable when fresh events can alter an immediate decision, delay creates business or safety risks, or users, devices, and teams need live coordination; batch suits deterministic, replayable, consistency-sensitive, and cost-intensive work such as reconciliation, training, reporting, and verification. The recommended production approach is usually hybrid: retain canonical business state in databases or durable event logs, stream small versioned notifications or state changes to affected participants, and use batch jobs for heavy transformations, audit, and recovery. Teams should assess each workflow step using factors including latency tolerance, staleness cost, fan-out, determinism, inference expense, consistency requirements, and operational readiness, then validate choices through SLOs, failure tests, cost measurement, and domain expertise. The discussion emphasizes idempotency, event versions, durable audit trails, scoped security, selective model invocation, and planned degraded behavior, while positioning PubNub for low-latency distribution and presence and Kafka or similar systems for durable replayable event processing.