Skip to content

Topic

AI agent harnesses

Software layers that wrap an AI model with tools, memory, prompts, and control loops so it can carry out multi-step tasks as an agent.

Current stories

buildOne report1 publisher

Pi 1.0 brings MCP into its core by batching tool calls in a JavaScript sandbox

Earendil's Pi 1.0, an agent harness whose website once advertised "No MCP", now builds MCP into its core. The team ties the reversal to Codemode, a JavaScript interpreter beside the harness loop that lets the model batch MCP calls into one script.

Publishers:dev.to

Reality

Evidence40
Adoption55
Hype gap+10
Incentives45
Confidence45
buildOne report1 publisher

MIT CSAIL's VISTA harness lifts Claude Opus 5.0 from 40.68 to 100 on ARC-AGI-3 efficiency

MIT CSAIL's VISTA harness raised Claude Opus 5.0's ARC-AGI-3 action-efficiency score from 40.68 to 100 without retraining or replacing the model. The comparison holds the model fixed, so the lift comes from the interface built around it.

Publishers:dev.to

Reality

Evidence38
Adoption
Not scored
Hype gap+40
Incentives
Not scored
Confidence35
productConfirmed5 publishers

Cohere pitches its North 2 agent platform to the administrator who has to cap the token bill

Cohere's North 2 agent platform lets administrators track token use down to each user and agent and cap spending across the whole company. Cohere is selling to whoever answers for agent cost and risk, and the launch report backs that with three customer accounts and no prices.

Perspective Coverage

5 publishers
Builder
Builder 31%
Operator
Operator 43%
Investor
Investor 26%

Reality

Evidence55
Adoption40
Hype gap+20
Incentives60
Confidence60
leadershipOne report1 publisher

Five harness rebuilds in two years anchor Jabali AI's case against waiting for better models

Jabali AI founder Vatsal Bhardwaj argues that the engineering around a model explains the 95% of enterprise AI pilots that MIT found showed no P&L impact. His five rebuilds in two years point to a standing budget, on evidence from one games company.

Publishers:forbes.com

Reality

Evidence25
Adoption
Not scored
Hype gap+30
Incentives70
Confidence30
productOne report1 publisher

SpaceX prices Grok 4.7 at $4.69 a task on a benchmark it owns

SpaceX says Grok 4.7 averages $4.69 per task on CursorBench 4.0, which was built by Cursor, one of its own acquisitions. The rivals it undercut were running quality-first configurations, and GPT-6 Astra beat it on chip design.

Publishers:siliconangle.com

Reality

Evidence28
Adoption16
Hype gap+45
Incentives80
Confidence58