AI Agents Can Go Off The Rails After 5 Or 50 Hidden Steps
AI agents can plan, use tools, remember context, and check results across many steps. This primer explains how they work and where they fail.
Aliasgar started writing after having come from the other side of the industry, spending six years building and supporting developer tools at protocols including Ignite, Informal Systems, Akash and Warden, much of it inside the Cosmos ecosystem. He holds a master's in computer science, has shipped products of his own, and writes from Germany.
AI agents can plan, use tools, remember context, and check results across many steps. This primer explains how they work and where they fail.
Ethereum Foundation sold 2,785 ETH, about $6.42 million, through a CoWSwap transaction that converted proceeds to USDS stablecoin, a routine treasury move
Oracle will invest $1 billion to expand incentives and resources for partners including cloud credits, co-selling funds and training stipends to accelerate
Mubit launched a platform that inserts a proxy URL between apps and model calls to train custom AI for repetitive office tasks, letting teams route requests
Cohere launched a public beta of Model Vault that runs managed model inference inside hardware‑isolated Microsoft Azure environments with NVIDIA confidential
OpenAI launched the Decisions API as a lower-cost, lower-latency endpoint for fast, focused model-driven yes/no choices, priority scores and routing that
PerturbAI and OpenAI announced a collaboration to apply multi-agent reasoning agents to large-scale biomedical data, chaining data ingestion, causal
Google showed Gemini 4 Argon turning fragmented archaeological records, raw LiDAR scans and archives from Chaco Canyon into an explorable 3D heritage site
Tepia, an AI systems integrator, was named an OpenAI Select Partner and will help businesses build and scale AI solutions across mobile, web and backend
A memory chip squeeze could slow AI, even after OpenAI's $250 billion cloud deal, because only a few suppliers make the memory every AI processor needs.