AI Application Architecture
Patterns for building production AI applications, including orchestration, memory, tool use, state management, APIs, and system design.
Gemini Managed Sandboxes Give Agents Persistent Cloud Environments
Philipp Schmid of Google DeepMind argues that AI agents should run in managed cloud sandboxes rather than on a developer’s machine. In his account of Google’s Gemini API, the Interactions API represents work as a sequence of steps, while the Antigravity agent runs code, uses tools and retains files in a persistent environment started with an API call. Schmid’s case is that Google can handle the agent loop and infrastructure, leaving developers to provide the request, tools and working context.
The Best Business Openings Start With Attention Others Overlook
Angel Studios founder Jeff Harmon argues that businesses can get their start by finding a “glitch”: an overlooked pocket of demand or attention that is small but motivated enough to support a foothold. In an interview with Shaan Puri, he points to Orabrush, where cheap early YouTube ads let the team test an offer and learn before the opportunity narrowed. His broader point is to pair a real customer problem with an underused way to reach people willing to act.
A Small Set of Choices Makes Model Decisions Nearly 10 Times Faster
OpenAI’s Decisions API is designed for applications that need a quick choice from a limited set of actions, rather than an open-ended response. In the launch demonstration, Romain Huet and Charlie Guo show how it turns text or image inputs into decisions an app can act on, such as routing sales leads or choosing a lane to steer a car. Huet says narrowing the model’s task to a few choices makes the API nearly 10 times faster.
Ecology Projects Combine Uploads, Real-Time Feeds, and Automated Model Analysis
The Sparrow Team at Microsoft’s AI for Good Lab presents Sparrow Studio as a workspace for bringing ecology data sources and analysis into a single project. Users can upload historical surveys or configure real-time ingestion from cameras and other devices, then assign models to analyze the data automatically. The tutorial’s central point is that each project ties together how data enters the system and how it is processed and managed.
The Scheduler Can Bottleneck Inference Without Doing the Heavy Compute
Charles Frye, who works on inference at Modal, argues that an inference engine’s central challenge is not generating tokens but serving them efficiently for the workload at hand. The scheduler can limit performance even though it does far less computation than the GPU, because it decides which requests reach the accelerator and how they are batched. Frye traces how workload demands, cache management and host-side overhead shape latency and throughput, and why operators need evidence from the deployed system to diagnose problems.
Coding Agents Need Testable Definitions of Done
Nick Nisi and Zack Proser of WorkOS argue that engineers can delegate more work to coding agents only when they define completion in terms the agents must verify. Their workshop turns a failed test loop—an agent touched a file meant to prove tests had run, without running them—into a broader operating rule: use measurable goals, automated checks and human review to keep parallel and recurring work in bounds.
Enterprises Are Building AI Strategies Around Model Independence
OpenRouter’s Alex Atallah and Replit CEO Amjad Masad argue that enterprise AI is likely to depend less on one all-purpose model and more on a mix of specialized systems that companies can choose, combine and control. Atallah makes the case for keeping model options open; Masad says businesses also need independence from model, cloud and data providers. Both see narrower models as potentially cheaper and easier to govern, though the speakers leave unresolved how agents should coordinate safely and when specialization beats a general model.
Pi Workspaces Turn Telegram Into a Gateway for Always-On AI Agents
Alejandro’s tutorial shows how to run specialized Pi assistants on an always-on machine and reach them through Telegram using his open-source `pi-gateway`. Each assistant runs from its own workspace, with its own instructions, tools and model; the gateway uses a local SQLite database to route messages to persistent Pi sessions. The demonstration covers a personal assistant and a research agent, while scheduled automations and support for other messaging platforms remain proposed extensions.
AI Agents Need Business Context to Run Enterprise Software Reliably
SAP chief executive Christian Klein argues that AI will change how enterprise software is used, not make systems such as SAP’s obsolete: agents still need reliable business data, process knowledge and controls to handle consequential work. In an interview with Big Technology, he says those requirements make accuracy and fragmented company systems key limits on adoption, even as he expects agents to take on more tasks. Klein also expects jobs to shift, with companies needing to reskill workers and change how they organize work.
AI Agents Write 741% More Code, but Review Limits What Ships
AI agents have sharply increased code output without a comparable rise in software shipped: developers using autonomous agents wrote 741% more code, but shipped only 30% more software. Laurie Voss, Arize AI’s head of developer relations and npm co-founder, argues that human review is now the bottleneck—and that asking people to review more code is not a scalable fix. The answer, she says, is to build automated review systems that test more than whether code passes, while keeping human judgment and production monitoring where automated checks fall short.
Eleven v4 Turbo Brings ~100-Millisecond Voice Responses to ElevenAgents
ElevenLabs says its v4 Turbo voice model, now available in ElevenAgents, combines a median inference latency of about 100 milliseconds with a wider emotional range and improved reliability on long calls. The company argues that those capabilities can help agents respond to sensitive customer needs with practical next steps, not just sympathetic language, while supporting conversations in more than 90 languages. Its demonstrations show agents offering a service-outage workaround, gathering details about a damaged order and presenting options for a parking citation.
Meta’s Muse Shows How Connectors Can Make AI Agents Useful
Meta’s Muse is being used to argue that consumer AI may depend less on the most powerful model than on software that connects to services and takes tedious tasks off users’ hands. On Big Technology, Alex Kantrowitz says Muse’s early use points to a product opportunity for Meta—and a challenge to intermediaries such as travel-booking sites—while Ranjan Roy cautions that adoption will hinge on trust and whether cheaper, non-frontier models can do enough. Neither sees Muse as proof that frontier AI has lost its value.
Task-Specific Detectors Can Beat Vision-Language Models at Runtime
Merve Noyan, a computer-vision and multimodal systems researcher at Hugging Face, argues that developers should stop deploying vision-language models as the default for vision tasks. They are too slow for many real-time uses, she says, and a task-specific detector can be faster and more accurate; licensing can also make popular model choices unsuitable. Her VibeVision toolkit instead gives coding agents task-specific models and a workflow that turns unlabeled images into a trained detector, with human checks on key decisions.
Data Agents Need Reviewed Semantic Context, Not Just Table Access
OpenAI argues that a useful data agent needs the same context a new analyst would receive, not just access to a warehouse. In its ChatGPT Work demonstration, the company shows how to combine Databricks tables, Slack discussions and metric documentation into a reusable semantic layer, then have a data expert review its definitions and rules before sharing it as a company plugin. The aim is to make the organization’s interpretation of metrics available alongside the underlying records.
Data Agents Need Governed Metric Definitions, Not Just Table Access
OpenAI’s data team argues that data agents need a governed semantic layer—not merely access to warehouse tables or SQL—to interpret metrics according to a company’s agreed definitions, populations, filters and source hierarchy. The team says organizations should reuse existing semantic work where possible, but have domain experts resolve contradictions and review the layer before distributing it as a shared skill. Maintaining that foundation is an ongoing governance responsibility, not a one-time plugin setup.
Hard-Tech Startups Rise as AI Lowers the Cost of Building
Y Combinator partners Garry Tan, Jared Friedman, Diana Hu and Harj Taggar argue that AI is reducing the software labor needed to build both physical and digital businesses, helping drive a rise in hard-tech startups and faster early revenue among YC companies. They say the opportunity is shifting toward companies that build physical capacity, own specialized data or take full responsibility for customer workflows—not software that merely records or assists work. AI also makes it easier for founders to start alone, they contend, while increasing the premium on judgment about what to build.
Natural Voice Agents Must Balance Latency, Intelligence, and Cost
Boson AI co-founder and chief executive Alex Smola argues that voice agents will not become useful simply by producing accurate speech: they must respond, pause, handle interruptions and use tools at the pace and emotional register of human conversation. That makes latency, audio representation and inference cost product constraints rather than back-end engineering details, he says. As systems add vision and avatar-like presence, Smola contends, they will need to manage those trade-offs while learning when and how to adapt to individual users.
Agent-Ready Interfaces Can Override a Decade of Brand Loyalty
Liad Yosef, co-creator and maintainer of the MCP Apps specification, argues that online services will increasingly compete on how readily agents can discover, authenticate with, and operate them—not just on the quality of their human-facing dashboards. He says companies should test agents completing real tasks rather than rely on published standards such as `llms.txt`, which Ora found agents often ignore. In Yosef’s model, services become “nearly headless”: agents handle routine operations directly, while providers return an interactive interface in chat when human judgment is required.
Full-Duplex Voice Agents Must Separate Conversation From Reasoning
Neil Zeghidour, co-founder and CEO of Gradium, argues that most real-time voice agents remain “walkie-talkies”: however quickly they respond, they can only listen or speak, not handle the overlapping signals that make human conversation work. Full-duplex speech models can represent simultaneous talk and backchanneling, he says, but they still sacrifice reasoning and tool-use capability relative to text-based agents. His proposed answer is a split architecture in which a lightweight speech model manages the live conversation while a separate text model handles harder reasoning and actions.
Voice Agents Should Act Through Interfaces, Not Just Speak
OpenAI developer-experience lead Charlie Guo argues that voice agents should not be designed to answer every spoken request with speech. The underused model, he says, is speech to action: a user states an intent, the system calls tools and changes the interface or completes a workflow, using audio only where conversation, hands-free operation or escalation warrants it. For developers, Guo’s practical case is that many of those tools already exist as application endpoints, handlers and workflows.
Managed Sandboxes Replace Agent Tool Registries With Skills and Filesystems
Google DeepMind’s Philipp Schmid argues that agent builders should stop encoding anticipated workflows in custom functions and instead give capable models a managed sandbox, general-purpose tools, clear instructions and tightly defined access. In his three versions of a GitHub pull-request reviewer, Python loops, tool schemas and registered functions give way to an `AGENTS.md` file, skill files and a remote environment that handles execution, state and credential injection. The work that remains with teams, he says, is setting behavior, permissions and evaluation—not maintaining orchestration that improving models can replace.
GPT-Live-1 Separates Live Voice From Agent Reasoning
OpenAI has released GPT-Live-1 in its API as a full-duplex voice frontend designed to keep spoken interactions moving through interruptions and background noise. The company positions the model as one layer of a broader application architecture: GPT-Live-1 handles live conversation, while a separately chosen backend provides reasoning, tool use and actions. Its listed price of 5 cents per minute applies only to that voice layer, not to backend inference or tool services.
Agents API Hosts Codex Orchestration for Production Incident Investigations
OpenAI’s Charlie Guo presents the Agents API as a hosted Codex harness for production agents, with OpenAI handling orchestration, sessions and context management while builders control instructions, tools, data access and execution environments. In a fictional checkout incident, the API is used to connect observability and code tools through MCP, filter large log datasets, run parallel investigation tasks and produce an evidence-backed handoff. The example is designed to recommend a mitigation with stated uncertainty, not to make a production change autonomously.
Longer Skills Files Shift Agent Design From Compression to Verification
Laurie Voss, head of developer relations at Arize AI, argues that frontier models’ capacity to retain simple prompt instructions has risen from roughly 200–300 constraints to thousands in about a year. Re-running and extending the IFScale benchmark, he found that the old compression problem has largely receded, but reliable compliance has not: models now fail through omissions, safety refusals, exhausted reasoning budgets or polished-looking partial answers. His practical conclusion is that teams should verify requirements in outputs rather than assume a fluent response followed the prompt.