Tag
42 entries tagged Agentic AI · 11 wires · 31 terms.
OpenAI published internal data on June 25 showing Codex at 85% of employee output tokens and 137x growth in non-developer use. The agent stack just became the default interface.
Google built computer use into Gemini 3.5 Flash, scoring 78.4 on OSWorld behind GPT-5.5 and Opus 4.8. The model now sees and operates your screen.
OpenAI and Broadcom unveiled Jalapeño, OpenAI's first in-house inference chip, design to tape-out in nine months, with gigawatt-scale deployment by end of 2026. The frontier lab is building its own foundation.
OpenAI ships Daybreak for the security pipeline, Anthropic launches Claude for Small Business with QuickBooks built in, and open-source OpenHuman builds a local Memory Tree. Three tools, one pattern.
OpenAI launched GPT-Realtime-2 plus voice translation and transcription. Same week, AI moved inside Excel, Word, and Acrobat. The tools are reshaping us.
OpenAI took ChatGPT ads to five new countries and opened a self-serve manager with CPC bidding. The free AI tool is now an ad-supported consumer product.
OpenAI made GPT-5.5 Instant the default for every ChatGPT user, claiming a 52.5% drop in hallucinations. The default still earns no automatic trust.
Cerebras prices 28M shares at $115-$125, targeting a $26.6B valuation on $510M revenue. Pentagon AI contracts freeze out Anthropic. Microsoft Agent 365 goes GA.
Google unveils eighth-gen TPU 8t/8i chips and the Gemini Enterprise Agent Platform at Cloud Next 2026, backed by a $750M partner fund. OpenAI, Tencent, Anthropic, and NEC round out a week of consolidation moves.
Amazon put up to $25B more into Anthropic. Anthropic pledged $100B back to AWS over a decade. The frontier lab is now a tenant of its largest investor.
Three tools that hand the craft over to an agent this week, and the warning that should land alongside them.
An open protocol that lets AI coding agents communicate with editors and IDEs in a standardized way, so the same agent can run inside Cursor, Devin Desktop, Zed, or any other compliant client.
A calibrated reliability rating attached to an AI agent's output, used to decide whether the result is safe to ship or must be routed to a human reviewer.
The user interface that lets a human watch, approve, and redirect an AI agent that is running somewhere else, while the underlying work stays on the machine where the agent lives.
A centralized directory that catalogs every AI agent operating within a system, assigns each one a verifiable identity, and tracks what each agent is authorized to do.
An agent swarm is a coordinated group of AI agents running simultaneously, each handling a sub-task, with results combined by an orchestrating system into a single output.
A web browser whose default surface is an AI agent that takes multi-step actions on the user's behalf across pages, not just a search box or a chatbot stapled to the sidebar.
The pattern where an AI agent, not a human, executes the search, decision, checkout, and after-sales steps of a purchase on a user's behalf.
The staffing pattern in which AI agents are counted alongside humans as productive units of an enterprise's labor capacity, with managers allocating work to whichever resource is faster, cheaper, or available.
An AI agent that runs persistently in the background of a workspace, watches the user's activity without being prompted, and takes initiative on what it sees.
A workflow in which an AI model both detects software vulnerabilities and generates the code fix, then opens or commits the patch with little or no human review in the loop.
An AI system sold as a unit of engineering labor rather than a coding assistant: you assign it a task and it plans, writes, tests, and submits the work for review on its own.
An AI agent that runs continuously on a remote server, executing tasks on a user's behalf even when the client device is closed.
An AI capability in which a model perceives a computer screen and operates its interface directly, clicking, typing, scrolling, and navigating apps to take actions on a user's behalf.
A foundation model purpose-built to operate software the way a person does, by clicking, typing, navigating menus, and manipulating files, rather than producing chat or code.
The practice of discovering, comparing, and buying products inside a conversational AI interface rather than on a traditional website or app.
The proportion of a team's output that is produced by an AI agent rather than by a human, measured per worker, per department, or per organization.
A JavaScript orchestration script that Claude writes and a runtime executes to fan a complex task across tens to hundreds of parallel subagents, keeping intermediate state outside the conversation.
The practice of AI capabilities moving into the response surface of existing productivity tools rather than living as a standalone application the user switches to.
A protocol extension that lets enterprise admins grant AI agents access to third-party connectors once through the company's identity provider, so individual employees inherit the access on first login instead of clicking through OAuth screens.
An attack that hides instructions inside content an AI agent reads, such as a web page or an email, so the agent executes them as if the user had given the command.
A chip designed from the ground up for one job, running already-trained AI models, rather than a general-purpose GPU adapted to that job.
An AI agent designed to sustain coherent action across many steps and long elapsed time without losing the thread of the task it was given.
An AI agent hosted by a platform (most commonly Anthropic's infrastructure) rather than running inside a personal terminal session. A Managed Agent persists beyond the operator's session, exposes a stable invocation surface (an API call or a button in an internal app), and reaches external systems through a governed tool gateway.
The background process by which an AI assistant rewrites and merges short-term context, saved facts, and prior chats into a smaller, cleaner, longer-lived store the model can use later.
A crafted input sequence that bypasses a model's safety guardrails and produces outputs the model was trained or filtered to refuse.
A credential, token, API key, service account, or OAuth grant issued to a software system, automation, or AI agent rather than to a human user, used to authenticate and authorize machine-to-machine actions inside enterprise environments.
AI configured to stay with one operator across all their tools and contexts, building a persistent memory of work patterns instead of starting fresh per session.
An AI skill creation pattern in which an agent watches a user complete a task once, then saves the demonstration as a reusable, parameterized routine the agent can execute later without further instruction.
The threshold at which an AI system can autonomously design, train, and ship a more capable version of itself, with each new generation authored by the previous one rather than by human engineers.
A saved instruction file that an AI tool reads each time you invoke it, producing consistent output across runs without requiring the operator to remember and retype a long prompt. In Claude Code, Skills live at ~/.claude/skills/<name>/SKILL.md and are invoked by typing /<name>.
An autonomous AI agent that a parent agent spawns to handle one bounded task, usually in parallel with sibling subagents, then return a structured result for the parent to synthesize.