Salesforce Updates Agentforce Vibes to Tackle AI Context Overload

Salesforce has updated its Agentforce Vibes coding platform to version 2.0, adding features designed to help enterprises control the amount of context fed to AI agents. The update comes as companies deploying coding agents report that excessive context is becoming a primary failure mode. Diego Mogollon, chief product officer at startup fundraising platform VentureCrowd, told VentureBeat that agents become "confidently wrong" when given too much unstructured data at runtime. VentureCrowd cut its front-end development cycle by 90 percent after deploying AI coding agents, but Mogollon said the challenges were "rarely about the coding agents themselves" and instead stemmed from messy data and unclear processes. The phenomenon, referred to as context bloat, occurs when AI systems accumulate more data, tools, and instructions than they can effectively process. The result is slower performance, higher token costs, and reduced reliability. Agentforce Vibes 2.0 introduces Abilities and Skills, which let enterprises define what agents should accomplish and which tools they should use. The platform also expanded support for third-party frameworks including ReAct. Unlike approaches that simply expand context as tasks evolve, Salesforce's method keeps context aligned with existing data models and codebases. Other platforms manage context differently. Claude Code and OpenAI's Codex use autonomous execution models that continuously read files and run commands, compacting context only when it grows too large. The consistent pattern across the market is that context keeps growing as workflows become more complex, making cost and latency harder to control. Mogollon said VentureCrowd chose Agentforce Vibes because its data already lives on Salesforce, and the platform allowed the company to control what context its agents receive.