Major AI tools added agents that do multi-step work on your computer and across apps. We also track new security offerings, pricing tiers, and government trust signals.
Several products moved from chat to hands-on help with multi-step work.
Perplexity was mainly a search-and-answer tool with sources. Now its “Computer” assistant carries out tasks for you, offered on the Max plan.
Perplexity Computer autonomous agent (19-model orchestration, Max tier)
Teams can shorten the gap between research and execution. Budgets may shift since the most useful workflow features sit behind a higher tier.
Perplexity's strengths centered on fast, cited answers. “Computer” now coordinates many AI engines and connects to hundreds of apps on the Max tier.
Perplexity Computer: autonomous agent orchestrating 19 models with 400+ app integrations (Max tier)
App connectivity often separates a demo from daily use. Test real tasks end-to-end and weigh the lock-in risk of tying workflows to one platform.
Cowork, Claude's PC task automation, is now available to paid users. A new Zoom connection pulls meeting information into Cowork.
Claude Cowork (PC-automation AI agent) now generally available for paid plan users; Zoom connector ingests meeting data into Cowork
Meeting notes can turn into follow-ups and action lists faster. It signals a push toward integrated work assistants rather than standalone chat.
Le Chat was primarily a conversational assistant. Agent Mode shifts it from “tell me” to running multi-step tasks more autonomously.
Le Chat gains Agent Mode for autonomous multi-step task execution
Evaluating Mistral now means workflow automation, not only model cost. Pilots should track time savings, error rates, and permitted actions.
Grok was best known for fast answers and X integration. “Grok Computer” now operates a desktop, planning and running steps in parallel.
Grok Computer: autonomous desktop automation agent with parallel planning and execution
Desktop automation can cut routine internal work. Check reliability, audit trails, and who can run actions before using it for critical tasks.
Creative work in Grok used to be prompt-driven and manual. “Imagine” adds a more autonomous mode for creative and image tasks.
Grok 4.3 "Imagine" agent mode: autonomous agent for creative and image-generation tasks
Marketing and design teams can get faster drafts and more variations per brief. Brand consistency and rights still need human review.
More AI features land directly in Word, team chat, and developer tooling.
Copilot already worked across Office apps. Word now includes a Legal Agent that flags clauses, checks for issues, and suggests redlines.
AI Legal Agent integrated into Microsoft Word for contract review, clause checking, and redline suggestions
Contract cycles can speed up as first-pass review gets faster. Define what needs human sign-off before relying on suggestions.
Poe was already a hub for trying many AI models. It now supports group chats for up to 200 participants, moving it toward a shared workspace.
Group chat for up to 200 users collaborating across 200+ AI models
Teams can evaluate models together with shared context. Plan access controls and guidelines for what gets shared in-room.
Claude Code is a terminal-based coding assistant. Uber's numbers shift it from “possible” to “proven at scale” in a large org.
Uber: ~95% of engineers use Claude Code monthly, ~70% of commits AI-generated (as of April 2026)
Strong adoption suggests measurable productivity gains. Review, testing, and standards need to keep pace—invest in training and guardrails, not just licenses.
Coding assistants were often framed as autocomplete. The updated guide highlights repo-wide understanding, multi-file changes, tests, and diff review.
Teams can plan for refactors, debugging, and test work. Evaluation shifts toward reliability with existing codebases, paired with verification habits.
Updates highlight vendor approaches to security, sensitive customers, and data.
Claude previously focused on general assistant and developer features. A dedicated security offering now gives the lineup a clearer “security buyer” story.
Claude Security launched: enterprise AI security tooling built on Mythos model capabilities, helping defenders gain AI advantage
Security teams may gain faster investigation and response workflows. Assess how it fits with existing security processes and reporting.
OpenAI already had broad model and API availability. This adds a major government selection signal for high-stakes use.
Selected for U.S. DoD classified AI program alongside Google, Nvidia, and xAI
It may de-risk vendor selection for sensitive workloads. Treat it as one data point—not a substitute for due diligence.
Claude is widely used for reasoning and coding. It was not selected for a U.S. classified AI program, with supply-chain risk cited.
Excluded from U.S. DoD classified AI program (cited as supply-chain risk; OpenAI, Google, Nvidia, xAI were selected)
Regulated sectors may ask tougher sourcing questions. Performance alone isn't deciding; map requirements to vendor capabilities.
ChatGPT's free tier has been moving toward ads in some countries. Marketing tracking is now on by default for free users there; paid plans are exempt.
Marketing cookies enabled by default for free users in ad-serving countries (paid plans exempt; opt-out available in settings)
Be cautious about employees using free accounts for work topics. Defaults matter because most users never change them.
Rapid progress in natural-sounding speech and medical benchmarks.
Gemini already covered a wide range of text tasks. A new preview adds expressive speech with emotion, pace, and style control, streaming with low delay.
Gemini 3.1 Flash TTS Preview: expressive speech synthesis with dynamic emotion/tempo/style control via Audio tags, low-latency streaming
Better voice quality helps call assistants, training, and accessibility. Low delay matters for live conversations, not just reading text aloud.
Medical AI claims are hard to compare across vendors. In a blinded physician simulation, DeepMind's system outperformed a leading alternative.
DeepMind AI Co-Clinician outperforms GPT-5.4 in blinded physician simulation test for medical diagnosis
It may accelerate pilots in triage support and documentation. Real-world deployment still requires validation, oversight, and accountability.
A few updates cover the basics: predictable pricing and reliability.
Grok's paid options ranged from bundled access to higher-priced tiers. A lower-cost SuperGrok Lite plan now sits between free and premium.
SuperGrok Lite: $10/mo (from 2026/3/25)
Lower entry pricing can expand pilots and small-team adoption. Compare what each tier includes before standardizing.
Early users saw degraded long-document performance in one distribution format. A third-party fix by Unsloth has restored it.
Mistral Medium 3.5 GGUF initial bug (YaRN parsing) fixed by Unsloth — long-context performance restored
Long documents are common in business work. Reliability fixes like this often matter more than new features when rolling into production.
A daily archive of changes actually applied to the site.
Google Home Speaker 2026 launches June 25 at $99 with Gemini LLM — Google's first speaker in six years. Frontier AI is now the default br…
Grok Gov confirmed inside U.S. targeting during Iran ops is the headline. Video AI price drops, morning briefings, and memory features al…
This morning's biggest moves: Cursor 3.7 and Claude Code Artifacts push agent tooling beyond the terminal. ChatGPT Enterprise also gets s…
Anthropic had a rough day: Fable 5 killed in 90 minutes, a Max plan lawsuit filed, and SDK billing reversed. SpaceX's $60B Cursor deal dr…
Salesforce's $3.6B Fin deal plugs the Agentforce inbound gap. In Japan, Firefly just turned a week of outsourced promos into a 20-minute …
Fable 5 went dark globally — a U.S. export order marks the first time a regulator has pulled a named frontier model. Fall back to Opus 4.…
Fable 5 is out — first Mythos-class Claude. The M&A story: Cohere bought Aleph Alpha, OpenAI acquired Ona for Codex, Reka merged with Moo…
Today's biggest story is OpenAI's $852B IPO filing. Anthropic's silent-nerfing apology dropped the same day — trust in both major AI labs…
Anthropic proposed FAA-style rules, the EU ordered Meta to open WhatsApp, and a whistleblower sued xAI — all in one day. DeepMind's Diffu…
Anthropic released Claude Fable 5, the first public Mythos-class model, across API, app, and Claude Code. Google shipped 70+ language liv…
Anthropic hired OpenAI's #2 custom chip engineer, a bet on owning inference silicon. Perplexity now lets AI write its search code, cuttin…
OpenAI added Lockdown Mode to ChatGPT as Meta confirmed thousands of AI-assisted Instagram takeovers. Trump's team is in talks over an Op…
Trump cut the U.S. AI review window from 90 to 30 days, widening the gap with the EU and Japan. Open models like Gemma 4 12B push frontie…
Anthropic filed confidentially for IPO days after its ~$1T Series H and opened Mythos broadly. NVIDIA pushed GPUs, AI PCs, and robots at …
US chip curbs now reach Chinese firms' overseas units, closing Southeast Asian bypasses. SoftBank's €75B French data center plan adds to …
Salesforce's first hard Claude Code numbers: a 231-day migration cut to 13 days. Codex gains autonomous Windows control; ChatGPT moves in…
OpenAI opened GPT-Rosalind free to vetted defense partners, with dual-use concerns raised. ChatGPT retires Canvas and o3/GPT-4.5; Nano Ba…
Claude Opus 4.8 is generally available with fast mode at one-third the price, and Anthropic's ~$65B Series H puts its valuation near $1T.…
Hassabis named 2029 as his AGI target, and Japan's three biggest SIs all aligned with Anthropic. Generative music and video posted real m…
AI agents posted hard numbers: near-2x on web tasks, 6.7x inference speed on 1T models. SDK vendors ship official Skills; hook-based enfo…
Two missing pieces for AI agent security landed at once. OWASP's Agentic Top 10 covers design reviews; Microsoft's RAMPART brings safety …
Gemini 3.5 Flash beats the 4-month-old Pro flagship at a third of the cost. Karpathy joins Anthropic the same day a 290,000-person Hitach…
Coding-agent competition shifted up a gear: Grok Build debuts, Codex goes on-prem via Dell, and Cursor ships Composer 2.5. Copilot Cowork…
Claude overtook ChatGPT in U.S. enterprise payments for the first time. OpenAI merged ChatGPT, Codex, and the API into one team, aiming f…
Two AI firms were hit by supply-chain attacks the same day — update macOS ChatGPT now; Mistral had ~450 repos stolen. Claude Pro splits a…
Manus went public and ChatGPT Agent replaced Operator, both with explicit pricing — agents are finally comparable. Devin 2.0 and device a…
AI moves from assistant to owner of the task. Claude now runs small-business back office, and Fanuc robots take plain-language orders via…
Claude overtook OpenAI in U.S. B2B adoption for the first time, plus small-business and legal launches. Google previewed Android automati…
GPT-5.5 Pro finished doctoral-level number theory research solo, verified by Fields Medalist Timothy Gowers. AI is shifting from assistan…
Anthropic shipped five announcements in one day, led by $30B ARR. Separately, ~5,000 vibe-coded apps (Lovable, Replit) were found leaking…
Anthropic doubled Claude Code rate limits on all paid plans after securing 220,000+ GPUs. A 2025 case adds a new risk: discipline for unv…
ChatGPT reveals which memories shaped each reply, Gemini 3.1 runs the smart home, and ElevenLabs hit an $11B valuation. A $200K prompt in…
Top model providers now pair their software with large-scale rollout services. Coding agents also moved closer to completing tickets end-…
Updates this week span trust, security, and creative workflows. One raises transparency concerns; others strengthen logins and speed up i…
Major AI tools added agents that do multi-step work on your computer and across apps. We also track new security offerings, pricing tiers…
AI-assisted coding gains transparency, model choices expand, and outputs export straight to business files. New guides cover AI run costs…
