GitHub showed both its scale and its strain
GitHub had the loudest fortnight because it occupied both sides of the ledger. On 23 Aug, it reported a seven-hour mass outage. On the same day, coverage noted that the platform now sees 2.9 billion commits a month, with commits doubling over the previous four months. On 23 Aug, GitHub's CTO publicly apologised to developers, with articles also referring to ongoing issues since April.
That is the core tension in AI coding infrastructure now. Usage is compounding, but reliability is becoming part of the product narrative rather than an operational footnote. On 22 Aug, GitHub published a blog post on the 17 Aug outage and future work. Over the same window, multiple articles covered Nintendo's takedown of more than 400 Switch emulator repositories hosted on GitHub, reminding the market that code hosting at this scale is also policy infrastructure.
The company still had momentum signals. On 23 Aug, GitHub added Zalando, Via, Tray.io, Sketch and Peak Money to its customers page, and published new case studies for Zalando, Via, Duolingo, Tray.io, Sketch, Peak Money, Netdata and Modsy. On 24 Aug, its total stars across top repositories rose from 19,644 to 56,837, and four repositories were created: awesome-copilot, emergency-pull-request-probot-app, dependabot-action and annotated-logger.
Cursor moved from editor to platform
Cursor's fortnight was not about another coding assistant feature. On 22 Aug and 23 Aug, multiple reports said SpaceX had completed its acquisition of Cursor for $60 billion. The same coverage noted that Cursor was outlining rebranding plans, including a reported identity shift to Ma.
The more strategically important move was Origin. On 22 Aug, multiple reports said Cursor had launched a code hosting service named Origin, available on GitHub, and positioned it as direct competition to GitHub. That timing matters. The launch landed in the same news cycle as GitHub outage coverage and developer frustration, which made code hosting feel newly contestable rather than settled.
Cursor's hiring also points beyond a narrow developer-tool posture. On 22 Aug, it had 113 job postings, including 48 in engineering, 30 in sales, 20 in other functions and 7 in marketing. That mix fits a company trying to turn heavy developer attention into a broader commercial system.
Lovable raised into scale, Replit priced usage down
Lovable's news was capital and usage. On 22 Aug and 23 Aug, it confirmed a $400 million Series C at a $13.3 billion valuation, with investors reported to include Tencent, the EU's Scaleup Fund and Menlo Ventures. The funding was described as supporting US expansion. The company also reported reaching 60 million projects, while a 23 Aug article said Lovable had reached 42 million visitors per month.
The point is not simply that investors are still funding AI software creation. It is that Lovable is being measured as a demand machine, not just a builder interface. The same funding coverage framed confidence in its potential beyond software development, which is the real market test for this category.
Replit attacked a different constraint: marginal usage. On 22 Aug, it introduced Free Mode using OpenAI's GPT-5.6 Luna, allowing users to perform unlimited lightweight AI tasks on paid plans. The launch appeared across seven articles. On 19 Aug, separate coverage said Replit had expanded access to its software creation capabilities by integrating GPT-5.6 Luna. Replit also had 71 active job postings on 22 Aug, with 36 in engineering, 13 in marketing and 7 in sales.
Vercel made trust and discoverability visible
Vercel used the fortnight to make two normally internal questions public: can agents understand your site, and can agents be safely contained. On 23 Aug, it launched Is Agentic, a free agent-readiness scoring tool that audits public websites with more than 100 checks. On 22 Aug, it offered $1 million in bounties for hackers who could escape its AI sandbox.
Those are not random launches. They put Vercel in the middle of how software is read by agents and how agentic execution is secured. On 22 Aug, Next.js 16.3 was also mentioned in coverage for instant navigations and reduced memory usage. On 23 Aug, Turborepo 2.10.11 was released, including automatic approval of release workflows, while Vercel's top repository stars rose from 207,397 to 239,350 and two repositories, swr and vercel-plugin, were added.
The quiet signal was commercial. On 22 Aug, Vercel had 83 job postings, all in sales and business development, including Account Executive and Business Development Representative roles across Commercial, Majors and Startups. In the same window, multiple articles discussed Vercel pricing in relation to Anthropic's costs, although no observed pricing change was recorded. Product surface, security posture and go-to-market capacity all moved in the same fortnight.
This was the week AI coding stopped looking like a feature race and started looking like an infrastructure fight. GitHub still showed formidable scale, with 2.9 billion commits a month and fresh enterprise proof on 23 Aug, but its outage made reliability part of the competitive conversation. Cursor used the same window to launch Origin and complete its $60 billion sale to SpaceX. Vercel pushed into agent readiness and sandbox assurance. Replit lowered the perceived cost of lightweight AI work, while Lovable raised $400 million into US expansion and mass usage. The market is widening from better code generation to ownership of hosting, trust, usage economics and distribution.
Each week this page takes a position and grades it in public once the horizon passes. Misses stay up. The full record.
This is the public read. OpsControl customers see this market live: every signal, graded and evidenced, the day it happens.
Track your own market
OpsControl