Calls

BTL3 agentic coding: 27B params, low overhead

BTL3 retains 92.2% of full-model behaviors in an 8.39 GB footprint, offering a specialized architecture for efficient agentic coding and tool use.

Jul 25, 2026 Marcus Chen 11 min read

Tool call requests: cut redundancy from 98% to 2%

Optimized frameworks slash redundant tool calls from 98% to 2%. Learn how structured schemas and event-driven resumption build reliable AI agents.

Jul 14, 2026 Sofia Berg 15 min read

Agent evaluation: Fix reasoning loops with glassbox traces

Move beyond static accuracy. Analyze multistep trajectories to catch critical failures where agents stall in continuous reasoning loops.

Jul 14, 2026 Marcus Chen 15 min read

SandboxRecord cuts latency in OpenHands cloud 1.38

OpenHands cloud 1.38.0 uses SandboxRecord to skip runtime API calls, removing network handshakes that slow down 40% of future agent apps.

Jun 23, 2026 Sofia Berg 13 min read

Autonomous agent loops burn $4,200 in a weekend

One developer lost $4,200 in a weekend. Learn why autonomous loops make 30–50 calls per ticket and how to set killswitches.

Jun 23, 2026 Sofia Berg 13 min read