Peter Farago

Where we build and version everything. RunLLM also connects directly to GitHub to help customers debug, document, and track issues. If there was an industry standard, this would be one of them. When someone say "PR" (Pull Request), Github is the word that comes to mind.

We use Slack more than just for communication on the team and with outside groups as well (24/7!). It's also part of our product experience where we ship, support customers, and run agents. Our agents post answers, escalations, and feedback in real time via Slack instances that our customers use to support their customers (a lot of companies use Slack as a technical support channel for their customers).

Our internal wiki, launch tracker, and buyer insights all live here. Flexible enough for content, structured enough for strategy. We use it as a company wiki, to coordinate tasks for group work, to write and give feedback on blog posts. Basically it's our knowledge base for everything non-technical.

This is our main product design tool, and our team lives in this. Our entire product UI (and launch visuals) were designed in Figma. Fast, collaborative, and essential for cross-functional design. Easy to share with outside groups too.

GCP gives us scalable infra with minimal friction. Their tooling around AI workloads has improved significantly in the past year. Easy integration with AI pipelines and generous scaling.

Gemini is arguably one of the top foundation models and with the most upside. They're executive incredibly well. Especially strong with coding and multi-modal inputs. A great model to keep in our multi-agent toolbox.

Claude 3 Opus handles longer context and nuanced conversation better than most. Great for logs, long-form reasoning, and fallback flows. Best long-context memory and thoughtful responses.

Still the most consistent model for high-stakes support answers. GPT-4-turbo gives us control, reliability, and depth — all critical for enterprise use. Most production-ready reasoning and tool use stability.

We use Amplitude to track how people engage with agents, what drives trust, and where friction lives. Great for product-led iteration. Clear funnels, good event modeling, fast insights.

It powers our multi-step workflows across agent actions. Temporal gives us durability, visibility, and precise control. Battle-tested orchestration with strong developer UX.

Modal helps us spin up Python workloads with speed and control. Great for agents that need to run fast, ephemeral jobs. A dream for short-lived compute without DevOps overhead.

The go-to dashboard for observability. We use it for real-time monitoring of agent reasoning, tool use, and feedback. Clear, customizable, and the fastest way to surface insights.

Time-series queries are fast, composable, and expressive. Perfect for support data, usage trends, and prompt analytics. Combines SQL flexibility with time-series performance.

Still the best all-purpose search engine when speed, flexibility, and integration matter. We use it heavily for vector + hybrid search. Scales better than most, works great with embeddings.

A joy to work with. Branching, schema changes, and zero-downtime deploys make PlanetScale perfect for our evolving data models. Developer-friendly + production-safe = rare combo.

It’s the backbone of our orchestration. K8s gives us control, isolation, and scalability across all our agent workloads. Still the best way to scale complex, stateful AI infra.

This makes creating a podcast very easy, even when participants are remote. Then post-production is also easy whether you're trying do some or all of it yourself or have a professional help. The ability to use AI to clean up pauses, pull out "ums" and other fillers, find the most useful marketing clips, are among some other cool features. Finally, sharing and collaboration features are also very good, as I use this within a team.

Fathom does a great job capturing strong meeting notes. I've used it to go through over a 100 sales call transcripts. I've also exported the transcripts (which isn't always easy to do in other solutions), and run them through ChatGPT to generate ICPs and Personas. So, I like it for what it can do for meetings, as well as making it easy for me to pull data out for different analyses elsewhere. Transcription seems very accurate, which is the key.

Notion has a lot of power, once you get used to using it. It's like a Wiki + task list + collaboration set of tools all at once. It's centralized work well for our company.