As of: 2026-11-13 13:49 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/v/claude-founders-managed-agents/ # How founders build on Claude Managed Agents Claude, 8 September 2026, YouTube. 117,089 views as of 8 November 2026. Kind: Interview. Watch: https://www.youtube.com/watch?v=hm8NzEd5io0 ## Why it is here Founders from Wispr, Actively and Pendo on building and scaling agents with Claude Managed Agents. _Verified 2026-09-29 from the YouTube watch page (title, channel, publish date 2026-09-08, length 34:42)._ ## Description (written by Gemini from the video) ### **Summary** This video features an Anthropic round-table discussion hosted by Lance Martin (Technical Staff at Anthropic) with startup founders Sahaj Garg (Co-Founder & CTO, Wispr Flow), Mihir Garimella (Co-Founder & CEO, Actively), and Todd Olson (Founder & CEO, Pendo). The panel explores how each company integrates Claude Managed Agents into their respective platforms, focusing on agent outcomes, organizational memory architectures, code sandboxing, evaluation strategies, and build-versus-buy trade-offs. --- ### **What is shown** * **[00:05]** Title card: *"How founders build on Claude Managed Agents"*. * **[00:15]** Sahaj Garg discusses using Managed Agents at Wispr Flow to automate meeting preparation (briefs) and post-meeting execution tasks. * **[00:52]** Mihir Garimella explains Actively’s model of running dedicated per-account sales agents alongside a cross-account intelligence product named "Watchtower." * **[01:35]** Todd Olson outlines Pendo’s agent integration, which inspects customer codebases against real user analytics in a sandbox to proactively suggest fixes and submit pull requests. * **[02:27]** Discussion on **Outcomes & Independent Verification**: Garg details how independent verifier agents with clean context windows evaluate briefs against a rubric before deciding whether to surface them to users. * **[07:04]** Discussion on **Agent Memory**: Garimella breaks down Actively’s dual-level memory model (persistent account-level agents vs. org-wide business logic and preferences). * **[10:48]** Discussion on **Sandboxing & Security**: Olson details sandboxing source code to safely inspect repositories, analyze telemetry, and generate pull requests. * **[12:20]** Discussion on **Build vs. Buy**: Panelists discuss why they chose managed agent harnesses over home-grown infrastructure during rapid iteration phases. * **[24:44]** Discussion on **Evals & Model Migrations**: Exploring early-stage "vibes-based" testing versus systematic evals, challenges of evaluating stateful memory and live third-party MCP tool calls (like Slack), and handling model style shifts. * **[28:57]** Discussion on **Cost & Platform Latency**: Requests for batch/flex modes to save 50–75% on offline tasks and pre-warmed sandboxes to reduce cold-start latency. --- ### **Claims & numbers** * **Mihir Garimella claims**: * Actively spun up their "Watchtower" cross-account product on Claude Managed Agents in about 2 weeks [01:27, 16:09]. * Running fan-out tasks across 500 accounts simultaneously makes top-tier frontier models too expensive without tiering to cheaper models [31:56]. * Adding batch or flex pricing modes would reduce offline background processing costs by 50% to 75% [32:36]. * **Todd Olson claims**: * Pendo spent time building custom agent infrastructure, encountered scaling and edge-case issues, and then migrated to Claude Managed Agents within two weeks [14:21, 22:21]. * Pendo had a working proof-of-concept running on Managed Agents in just a few days [14:50]. * **Sahaj Garg claims**: * Wispr Flow was able to build the first version of their meeting preparation feature in a single day using Managed Agents [15:04]. * Over a few weeks, Wispr Flow scaled up their user base by 100x to 1000x on Managed Agents with only a few days of iteration on system logic [15:08]. * Wispr Flow runs pre-meeting brief preparation agents roughly 24 hours prior to scheduled meetings [05:27]. --- ### **Notable quotes** 1. **Sahaj Garg [03:01]:** *"Being able to correctly identify whether the agent produced the right outcome, and literally choose not to show the user anything at all if it didn't, is way better than giving the user a false positive information."* 2. **Todd Olson [13:38]:** *"You don't need to roll your own infrastructure to solve those problems... None of us in this round table, we're not infrastructure folks."* 3. **Mihir Garimella [28:34]:** *"If a new model introduces new failure modes that are specific to it that you have to avoid, that's probably the most valuable to us."* --- ### **Assessment** This is an official promotional fireside discussion produced by Anthropic showcasing founder case studies for Claude Managed Agents. The video consists of candid technical discussions and architecture explanations without live screen captures or on-screen code demonstrations. _Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames._ ## Related - 2026-04-08: [Anthropic launches Claude Managed Agents (public beta)](https://postcutoff.com/e/2026-04-08-claude-managed-agents/) What else happened that month, day by day: https://postcutoff.com/timeline/2026/04/