- Why review matters more in the age of AI coding: as teams generate more code with agents, the main constraint becomes verification, not output, especially for critical or complex changes
- Propel’s core idea: embedding AI review into pull requests and agent workflows so teams can automate low-risk review work, preserve human attention for higher-stakes decisions, and build feedback loops over time
- What they demoed + discussed: AI-reviewed PRs, configurable approval thresholds by risk area, repo-readiness analysis, and continuous review loops that help teams shift verification earlier in the development process

D2 Product Preview | Propel Code: When Agents Need a Second Opinion - Building Review Into Agentic Workflows

- Why SDKs are now central to API quality: most developers use APIs through SDKs rather than raw HTTP, so type safety, ergonomics, and documentation increasingly define the overall developer experience
- Stainless’s core idea: using OpenAPI plus lightweight configuration to generate high-quality SDKs, docs, CLIs, and MCP servers, helping teams deliver a more polished, scalable, and Stripe-level API experience
- What they demoed + discussed: generating multi-language SDKs from a single spec, layering custom functionality on top of generated clients, and how modern API tooling is evolving to better support AI agents, developer workflows, and long-term maintainability

D2 Product Preview | Stainless: How to build world-class APIs

- Why multimodal is a shift: video, images, and audio are becoming first-class AI data, but most stacks still treat them as opaque files, leading to brittle pipelines and high maintenance overhead 
- Pixeltable’s core idea: a more “database-like,” declarative table/view model with native media types, built-in transforms + model inference, and strong lineage/observability 
- What we demoed + discussed: ingesting video by URL, extracting/transcribing/chunking, building embeddings for semantic search, and an auto-crop example, plus Q&A on flexibility (UDFs/models), private deployment, and scaling direction

D2 Product Preview | Pixeltable: The Multimodal Shift - Rethinking Data Infra for the Age of Vision, A/V & LLMs

In this session, we explored why agentic AI can be risky: models can sound confident while hallucinating, skipping required steps, or overstepping their authority. We covered semantic authorization, granting permissions based on meaning, context, and output quality, and demoed a control-plane pattern that enforces pre-checks, calibrates thresholds against known-good responses, and records a full audit trail tied to a human owner for real workflows.

D2 Product Preview | Manual Software: Agentic Safety, Semantic Authorization & Guardrails for Real-World AI Actions

Models and agents are actually the commodities; your business context is the moat.Model providers are constantly one-upping each other: Gemini 3 was state of the art...for about a week. GPT–5.3 blew it out of the water, only to be topped by Sonnet 4.6 shortly after.Switching costs are still high. How can you ensure your team isn't stuck on the same model forever?Winning teams are the ones who maintain sovereignty over their business context and institutional knowledge, allowing you to swap out models and tools on a dime.Don't just give your institutional knowledge away to the model providers to train on: own your context to maintain control and apply it on your own terms.In this D2 Product Preview, Dave Nunez (Falconer Co-Founder &amp; CEO) will dig into how to build context that powers whatever model or agent comes next, without hitting reset every time the landscape shifts. He'll demo how <a href="https://falconer.com/" rel="noopener noreferrer" target="_blank">Falconer</a> helps teams build their context graph, then open the floor for conversation on how to make context compound instead of disappear.

D2 Product Preview | Context is the moat: why your knowledge layer outlasts every model

AI coding tools have transformed how fast teams ship—but speed without oversight creates a new kind of risk. As agentic workflows take over more of the development cycle, many teams are discovering a quiet problem: when the same model that writes your code also reviews it, you're not getting a second opinion. You're getting confirmation bias at scale.Join this roundtable to explore what independent AI code review actually looks like in practice. We'll hear from Tony Dong, Founder &amp; CEO at Propel Code, who will share a brief overview of their multi-model approach to code review—built specifically to be independent from code generation, integrated directly into agentic workflows, and customizable down to the team, repo, and developer level.Then we'll open it up to the group:<ol><li data-list="bullet">How are you maintaining quality and standards as AI-generated code volumes increase?</li><li data-list="bullet">Where does institutional knowledge live today—and how do you make sure AI has access to it at review time?</li><li data-list="bullet">How do you decide when human judgment needs to be in the loop?</li></ol>Whether you're running a single coding agent or a fleet of them, this is a peer conversation for engineering leaders navigating the gap between shipping faster and shipping better.

D2 Product Preview | When Agents Need a Second Opinion: Building Review Into Agentic Workflows

As AI agents take on more and more work for us, they will need consistent, well-documented, programmable interfaces to accomplish the work we need them to do. In this world, every company must become an API company. But how do you build a truly great API, both for agents and human developers? In this session, we'll cover the components of a successful API program, from API design and development, to rate limiting and debugging, all the way to delivering last-mile developer and agent interfaces like SDKs, docs, and MCP servers. You'll learn how API teams at Stripe, Anthropic, and more scaled up to become foundational API primitives of the Internet, and set a new standard for developer experience. You'll leave with an operational blueprint of a high-functioning API program at scale, one that ensures both agents and human integrators are consistently successful.

D2 Product Preview | How to build world-class APIs

As AI applications increasingly combine vision, audio, video, and LLMs, teams are drowning in fragmented infrastructure. A typical production AI workflow requires stitching together 5+ services: cloud storage for raw files, a separate metadata database, an orchestrator like Airflow, embedding services, vector databases for similarity search, and hundreds of lines of glue code to keep everything in sync. Every time a model improves or business requirements change, you're rewriting pipelines and re-processing entire datasets. This isn't building AI - it's maintaining infrastructure.In this session, you'll see how a declarative approach to AI data infrastructure eliminates the glue code and maintenance burden. Instead of writing pipelines, you define transformations as computed columns and tablesAnd we'll walk through a real multimodal workflow, e.g. starting with raw video, extracting audio, transcribing with Whisper, generating embeddings with CLIP, and building semantic search. Then we'll show what happens when you update upstream data and/or add new transformations.This session is most relevant for ML Engineers, AI App Developers, and technical leaders who are:<ol><li data-list="bullet">Building production AI applications that combine images, video, audio, or documents with LLMs</li><li data-list="bullet">Currently managing multiple services (S3 + Postgres + Airflow + Vector DB + custom scripts)</li><li data-list="bullet">Hitting scaling pain points: slow iteration, complex model updates, or lack of reproducibility</li><li data-list="bullet">Looking to reduce infrastructure complexity without sacrificing capability</li></ol>

D2 Product Preview | The Multimodal Shift - Rethinking Data Infra for the Age of Vision, A/V & LLMs

As more teams move beyond chat over data and start giving agentic systems the ability to&nbsp;take real actions—triggering workflows, changing state, even touching money—the old safety patterns start to crack. Role-based access control on APIs and a few regex-based guardrails on model output aren’t enough when you have autonomous or tool-using agents making decisions inside complex, ambiguous business processes. Agentic systems today are substantially limited by hallucination, adherence problems, poor context length, increasingly high token use, and model attention.In this roundtable, we’ll explore how to make agents significantly more effective, using emerging patterns for&nbsp;agentic safety: how we direct and constrain what agents are allowed to do, on whose behalf, and under what conditions. We’ll focus on&nbsp;semantic authorization—authorizing “what this action means in the business context” rather than just “which endpoint is called”—and on practical, layered&nbsp;guardrails&nbsp;that go beyond prompt engineering or AI red-teaming.Chad Wathington is the CEO of Manual Software. His career has spanned product, strategy, and engineering leadership roles. Some of you may have attended his session @ ELC Annual 2024. He will give a brief overview and demo of a semantic authorization control plane, and then the group will spend most of the time comparing notes:<ol><li data-list="bullet">How are you preventing unsafe or unintended agent actions today?</li><li data-list="bullet">Where do current tools and patterns fall short for your domain?</li><li data-list="bullet">What would a credible safety, governance, and audit story look like for agents in your organization?</li></ol>

D2 Product Preview | Agentic Safety, Semantic Authorization & Guardrails for Real-World AI Actions

Exclusively for D2 members to connect, collaborate, and grow together.

D2 Connect+

Home

Forum

Events

Content

People

Clubs

Chat

Channels

Tools

Help

D2 Connect

Login to the D2 Community to connect with others, attend community events, and more!

Login or Sign Up for even more

To unlock all parts of the community and get the best experience, complete your profile.