Contrary to the trend toward multi-agent systems, Tasklet finds that one powerful agent with access to all context and tools is superior for a single user's goals. Splitting tasks among specialized agents is less effective than giving one generalist agent all information, as foundation models are already experts at everything.
Related Insights
Multi-agent systems work well for easily parallelizable, "read-only" tasks like research, where sub-agents gather context independently. They are much trickier for "write" tasks like coding, where conflicting decisions between agents create integration problems.
For complex, multi-turn agentic workflows, Tasklet prioritizes a model's iterative performance over standard benchmarks. Anthropic's models are chosen based on a qualitative "vibe" of being superior over long sequences of tool use, a nuance that quantitative evaluations often miss.
Instead of one monolithic agent, build a multi-agent system. Start with a simple classifier agent to determine user intent (e.g., sales vs. support). Then, route the request to a different, specialized agent trained for that specific task. This architecture improves accuracy, efficiency, and simplifies development.
To avoid confusing agents with contradictory goals, Tasklet plans to shift from pre-generated, static instructions to dynamically generating them just-in-time for each task run. This ensures the agent always operates on the most current user feedback, preventing errors from conflicting historical directives.
When building Spiral, a single large language model trying to both interview the user and write content failed due to "context rot." The solution was a multi-agent system where an "interviewer" agent hands off the full context to a separate "writer" agent, improving performance and reliability.
Tasklet's CEO argues that while traditional workflow automation seems safer, agentic systems that let the model plan and execute will ultimately prove more robust. They can handle unexpected errors and nuance that break rigid, pre-defined workflows, a bet on future model improvements.
Building a single, all-purpose AI is like hiring one person for every company role. To maximize accuracy and creativity, build multiple custom GPTs, each trained for a specific function like copywriting or operations, and have them collaborate.
Don't fear deploying a specialized, multi-agent customer experience. Even if a customer interacts with several different AI agents, it's superior to being bounced between human agents who lose context. Each AI agent can retain the full conversation history, providing a more coherent and efficient experience.
Tasklet, a platform for automating recurring tasks, found a surprising user behavior: most messages are for ad-hoc, one-off requests. Users invest time creating a highly-contextualized agent for automation, then leverage that same smart agent for immediate, chat-based assistance, making chat the dominant interaction model.
To make agents useful over long periods, Tasklet engineers an "illusion" of infinite memory. Instead of feeding a long chat history, they use advanced context engineering: LLM-based compaction, scoping context for sub-agents, and having the LLM manage its own state in a SQL database to recall relevant information efficiently.