3 Founders Shipped Claude Managed…

1. Key Themes

Build vs. Buy Is Now a Single, Repeatable Test

All three founders converged on identical logic despite building different products: infrastructure should be bought unless it is the differentiator.

"It really comes back to if you don't need to roll your own infrastructure for it because it's not the thing that defines the quality of what you're building, choose to buy, and in the cases where it is, build it yourself."
"Buy the stack when it's a commodity. Build it when it's the product. Every founder at this table proved both halves inside the same year."

Speed Is Now Measured in Days, Not Quarters — But Only for Infrastructure

Once the infra decision is made correctly, iteration speed compounds dramatically because engineering time redirects entirely to product validation.

"It allowed us to build the first version of the feature in a day and then ramp up with like 100 to 1,000x scale up in the user base over the course of a few weeks."
"So the time went where it should: figuring out what users needed, testing against real usage, repeating. Once infrastructure is solved, speed is iteration cycles, never headcount."

Independent Verification Beats Self-Review in Agentic Systems

A recurring architectural pattern: a second, context-isolated agent grades the first agent's output, since self-assessment is structurally unreliable.

"The fully independent clean context window means that you can assess the quality of the work as part of deploying it."
"An agent reviewing its own trace will pass itself every time. Hand the grade to a stranger and the rubric starts meaning something."

Memory Architecture Needs Explicit Scope Separation

Teams that blurred account-level vs. cross-account memory paid for it later; this is emerging as a core design discipline for agentic products.

"There's memory on an account and across accounts. There's memory on a user level and an org level."
"Mix the two and everything works fine for months. Then three engineers spend a quarter untangling it."

Evaluation Remains the Unsolved Frontier

Despite production maturity elsewhere, no team has cracked evals for stateful, memory-driven agents — this is presented as an industry-wide, not company-specific, gap.

"I don't think we found a good way to eval given memory. A system that's very stateful with this memory that is live."
"Phase zero is vibes-based."

2. Contrarian Perspectives

Chasing frontier model performance is a distraction, not an advantage. Rather than racing to adopt the newest model for marginal capability gains, the smarter posture is defensive — hunting for new failure modes the upgrade introduces.

"Skip chasing the last 1% of performance, each generation hands you that for free. Spend the energy catching failure modes specific to the new model... before a live user does."
Evidence: "The whole five series of models tends to produce writing in a way that has actually a lot more AI telltale signs than previous series of models."

Cost optimization early is a "well-organized distraction," not a virtue. Against the instinct to optimize spend early, the founders argue premature cost engineering delays shipping without proportional benefit.

"Cost work pays off the day it makes something buildable that wasn't. Before that day it's a well-organized distraction from shipping."

A wrong answer can be worse than no answer at all — silence is a legitimate product feature. Counter to the "always show something" instinct in AI UX, this team chose to suppress output entirely rather than risk confidently wrong personalization.

"If you're talking to a person named Tom and it gives you information about the wrong Tom, it's actually almost destructive."
"The fix: a second agent that grades every brief before a user sees it. Fail the grade, show nothing. Silence over confident error."

3. Companies Identified

Oracle — Cloud/enterprise infrastructure provider; sponsor content on agent memory. Mentioned as a case study for solving agent memory decay via a structured, four-type memory system.

"They gave agents four kinds of memory: working, semantic, episodic, and procedural. In their tests, agents scored almost 94% on LongMemEval and used a tenth of the tokens after 80 turns."

Anthropic (Claude / Claude Managed Agents) — AI model and agent infrastructure provider. The underlying platform enabling all three founders' rapid builds; central subject of the article.

"Claude Managed Agents hit the scene saying their bots could run by themselves, no babysitting needed."

Saj's company (unnamed, meeting assistant) — AI meeting brief/assistant product. Case study for the self-grading agent pattern and rapid one-day-to-1000x scaling.

"It allowed us to build the first version of the feature in a day and then ramp up with like 100 to 1,000x scale up in the user base over the course of a few weeks."

Mahir's company (unnamed, sales agent platform / "Watchtower") — Sales agent fleet + cross-account intelligence layer. Case study for scaling a fleet of per-account agents and building a cross-cutting feature ("Watchtower") in two weeks.

"We spun that up on managed agents in about 2 weeks. It's now one of our most used features, and it powers a lot of other parts of the product."

Todd's company (unnamed, code-review/PR agent) — Agent that reads customer codebases and submits pull requests. Case study for security/sandboxing risk and a build-vs-buy reversal after hitting infrastructure limits.

"You can imagine that people interacting with an agent, and you get access to someone's source code, you can potentially get secrets."

4. People Identified

Saj — Founder building a meeting-assistant product. Cited for the "wrong Tom" failure story and the self-grading agent design; also for the fastest build-to-scale example in the piece.

"If you're talking to a person named Tom and it gives you information about the wrong Tom, it's actually almost destructive."

Mahir — Founder building a sales agent fleet product. Cited for shipping the Watchtower cross-account feature in two weeks and articulating the account vs. cross-account memory distinction.

"There's memory on an account and across accounts. There's memory on a user level and an org level."

Todd — Founder building a code-analysis/PR-generating agent. Cited for security concerns around source code access and for reversing an initial build decision in favor of a managed sandbox solution.

"Todd came from the other side: homegrown solution, wall, proper bake-off against a managed option, then the rebuild decision."

Ruben Dominguez — Author of the newsletter piece. Conducted the interviews and synthesized the cross-founder playbook.

(Byline; conducted the reporting behind all quotes in the piece.)

5. Operating Insights

Ship narrow, let cross-cutting products emerge organically. "2 of the 3 products here started as one feature and became a second product only when a cross-cutting need showed up on its own. Ship narrow. Let the second product earn its place."
Always pair an agent-graded output with an independent, context-isolated grader before shipping anything user-facing. "Copy the outcomes pattern before you ship anything agent-graded."
Use four questions to decide build vs. buy: control, cost, time, and whether you're still validating the idea works at all — a concrete, repeatable framework rather than a gut call.

6. Overlooked Insights

Batch processing is a major untapped cost lever specifically for time-shiftable agent workloads, yet remains largely unbuilt across the tooling ecosystem: "If we're running things 24 hours in advance, we could probably save 50% to 75% on cost and make that possible... It mostly doesn't exist yet."
Eval suites can silently rot when over-fit to a prior model generation's quirks, meaning teams may believe they have coverage when they're actually "testing nostalgia" — a subtle but consequential blind spot as model upgrades roll out: "A good eval suite catches most of it, on one condition: the prompts weren't over-tuned to the previous model family's quirks. Otherwise you're testing nostalgia."