Replit launched Missions on Tuesday, 17 November, a mode in which a “swarm of agents” works toward an outcome the user defines, within a spending limit and approval rules set upfront, according to the company’s launch post. In its own test, Replit said a Missions build took in application traces at 49,000 per second, against 9,800 for a monitoring platform built from the same prompt with /goal in Anthropic’s Claude Code: “5× faster”.
Our number: in Replit’s NumPy benchmark, doubling the swarm from 16 to 32 agents lifted the speedup from 17x to 19x, about 12% more, by our arithmetic, while going from one agent to 16 lifted it from 4x to 17x. The $3,000 budget Replit gave its own test mission is 30 times the $100 monthly price of its Pro plan, by our arithmetic.
Business Pill · Usage-Based Pricing
An evergreen Business Engineer explainer on usage-based pricing, paying for work rather than seats. It teaches the general idea only and does not discuss Replit or Missions.
The key insight: Replit is selling agent work as a budget-capped outcome rather than a seat: a Missions agreement sets the spending limit, and an AI judge checks results against agreed criteria. In its own NumPy test, going from 16 to 32 agents added about 12% more speedup, by our arithmetic.
What Replit Launched
Replit said users “define the outcome” and Missions “coordinates a swarm of agents to plan, build, test, and iterate toward it within the budget and boundaries you set”. During setup, a Missions agreement sets the spending limit and “which actions require your permission”.
An “independent AI judge” checks results against the agreed success criteria, the company said. “If the budget runs out first, work stops with the outcome still unresolved”.
Replit’s example goals include “win our first 10 paying enterprise customers”, with approval required before the agent contacts prospects or shares a proposal. Users start one by clicking “New” in the sidebar and selecting “Mission”, the post said.

The DataCat Test
Replit said it gave Missions the goal of building DataCat, its own application monitoring platform. “In 2 days, with a $3,000 budget”, the mission built dashboards, alerts and application tracing, the company wrote, and it is open-sourcing a self-hostable version.
- Failure detection. DataCat raised an alert on a deliberately introduced database bug in 38 seconds, against a 60-second goal, Replit said.
- Load test. It stored 26 million activity records and 4.4 million log entries over three hours “with zero data loss”, and passed 51 of 51 data and alert checks, per Replit.
- Against Claude Code. Replit reported 49,000 traces per second against 9,800, 405,100 metric points per second against 224,600, and 51 of 56 API compatibility checks against 45 of 56 for the Claude Code build.
Replit also wrote that “The Claude Code build covered more product areas and took in logs faster”, and that the build comparison “used one run per setup on identical cloud machines”.
How Far the Swarm Scales
On 13 NumPy benchmark tasks, Replit said, a single agent on Auto (Max) reached speedups of up to 4x, a 16-agent swarm 17x and a 32-agent swarm 19x. On Auto (Power) the same steps gave 2x, 7x and 9x. Most of the gain came in the first 16 agents in both settings, by our reading of Replit’s figures; our explainer on agent swarms covers how multi-agent systems split work.
The Business Lever
The lever is spend per outcome rather than per seat. Replit’s FAQ answers “What does it cost, and which plans include it?” with “Set a maximum budget for your mission upfront. Work performed during the mission counts toward that budget”. Its pricing page lists Core at $20 a month and Pro at $100 a month with $100 of credits, with higher Pro tiers up to $2,000 a month.
Replit’s CEO earlier said the company moved off Anthropic’s models on cost and performance. Missions compares itself directly with Claude Code, the product at the centre of today’s developer debate over GPT-6.2 Astra.
The Structural Read
As we read the launch post, every performance figure, including the 5x trace-intake result against a Claude Code build, comes from Replit’s own tests, and Replit reported areas where the Claude Code build did better.
As we read the NumPy figures, the step from one agent to 16 lifted the Auto (Max) speedup from 4x to 17x, while the next 16 agents added 2 more points, by our arithmetic.
Replit, launch post, 17 November 2026
“If the budget runs out first, work stops with the outcome still unresolved”
Three Implications
FOR BUYERS Replit said users set a spending limit and decide which actions need approval before a mission starts.
FOR REPLIT Its pricing page lists Pro at $100 a month with $100 of credits; the $3,000 DataCat budget is 30 times that, by our arithmetic.
FOR ANTHROPIC Replit compared a Missions build with one made using /goal in Claude Code and reported both wins and losses for each.
The Business Engineer Lens
This story maps onto the Business Engineer framework The Agentic Harness War.
The framework says: “Foundation models become relatively commoditized as orchestration layers capture differentiation.”
As we read it, Missions puts Replit in the orchestration layer: coordinating many agents, budgets and an AI judge, while comparing itself with Claude Code. The framework predates this launch and does not discuss Replit or Missions.
What Is Not Established
Every benchmark here is Replit’s own, run on its own tasks; Replit wrote that its controlled tests “do not establish performance against the original platform’s hosted service or on Replit’s production systems”. It did not name the original monitoring platform. The Missions page does not say which plans include Missions or publish a per-mission price, and Replit did not give a launch-day user count. The 12%, the 30x and the reading of where the gain comes from are our arithmetic and interpretation. We did not contact Replit or Anthropic.
The Bottom Line
Replit launched Missions, budget-capped agent swarms judged against agreed success criteria, and said a $3,000 mission built a monitoring platform that took in traces 5x faster than a Claude Code build. In its own NumPy test, going from 16 to 32 agents added about 12% more speedup, by our arithmetic.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The launch details, quotations and every benchmark figure are from Replit’s launch post of 17 November 2026, its Missions product page and its pricing page, read in full on replit.com that evening; the benchmarks are Replit’s own and were not independently verified. The comparison with Claude Code is Replit’s. The roughly 12% gain and the 30x ratio are our arithmetic. We did not contact Replit or Anthropic.
Sources: Replit, Introducing Replit Missions, 17 Nov 2026 · Replit Missions product page and FAQ · Replit pricing page









