Ramp and Replit both say anyone can ship code now. Amazon says stop counting it
Two companies shipped the same idea within a few hours on Tuesday: a cloud coding agent that lets anyone in the company, not just engineers, hand over a tested pull request.
Ramp opened early access to Inspect, the coding agent it already uses internally. The numbers in CTO Karim Atiyeh's post are big. In the past month 99% of Ramp's engineering, design and product teams used it, Inspect "now accounts for 74% of merged PRs," and Ramp's builders ship 6x as much code per person as a year ago. You tag it from Slack, GitHub or Linear, every session is multiplayer, and it returns "a visually validated and fully reviewed pull request." Replit made Cloud Agents generally available for Pro and Enterprise customers the same morning: the agent opens the running app in a browser, clicks through the change, and attaches screenshots and video. Replit says a workspace memory that turns review comments into skills engineers approve cut its own token use by 51%.
Both products answer the same question: once code is cheap, what does the human reviewer need? Their answer is evidence attached to the PR, so the engineer reviews instead of reproduces. Ramp's post says it directly: the bottleneck is "increasingly, it's our own judgment."
Amazon picked the same day to publish its internal AI engineering guidance, and it reads like a reply. Per CNBC, the memo calls token usage, adoption rate and code-change counts "deeply flawed" measures and says to judge teams by long-term outcomes. One line deserves to travel: "If your work does not benefit from current AI tooling, that is a finding, not a failing." Amazon shut down its internal token leaderboard in May.
Both can be true. 74% of merged PRs is a real adoption signal, but it is not an outcome. The teams that win with these tools will be the ones that measure what reached customers and what broke, not how many PRs the agent opened.
Ramp Inspect: https://ramp.com/blog/introducing-inspect Replit Cloud Agents: https://replit.com/blog/anyone-can-bring-ideas-to-company-codebase Amazon memo coverage: https://www.cnbc.com/2026/10/27/amazon-ai-coding-tokenmaxxing.html
← Back to all articles
Ramp opened early access to Inspect, the coding agent it already uses internally. The numbers in CTO Karim Atiyeh's post are big. In the past month 99% of Ramp's engineering, design and product teams used it, Inspect "now accounts for 74% of merged PRs," and Ramp's builders ship 6x as much code per person as a year ago. You tag it from Slack, GitHub or Linear, every session is multiplayer, and it returns "a visually validated and fully reviewed pull request." Replit made Cloud Agents generally available for Pro and Enterprise customers the same morning: the agent opens the running app in a browser, clicks through the change, and attaches screenshots and video. Replit says a workspace memory that turns review comments into skills engineers approve cut its own token use by 51%.
Both products answer the same question: once code is cheap, what does the human reviewer need? Their answer is evidence attached to the PR, so the engineer reviews instead of reproduces. Ramp's post says it directly: the bottleneck is "increasingly, it's our own judgment."
Amazon picked the same day to publish its internal AI engineering guidance, and it reads like a reply. Per CNBC, the memo calls token usage, adoption rate and code-change counts "deeply flawed" measures and says to judge teams by long-term outcomes. One line deserves to travel: "If your work does not benefit from current AI tooling, that is a finding, not a failing." Amazon shut down its internal token leaderboard in May.
Both can be true. 74% of merged PRs is a real adoption signal, but it is not an outcome. The teams that win with these tools will be the ones that measure what reached customers and what broke, not how many PRs the agent opened.
Ramp Inspect: https://ramp.com/blog/introducing-inspect Replit Cloud Agents: https://replit.com/blog/anyone-can-bring-ideas-to-company-codebase Amazon memo coverage: https://www.cnbc.com/2026/10/27/amazon-ai-coding-tokenmaxxing.html
Comments