Factory's $50M bet on "agnostic" AI coding agents arrives as the market consolidates around single vendors. With Nvidia and J.P. Morgan backing their multi-LLM approach, the startup claims enterprise gains that challenge typical adoption curves.
Good Morning from San Francisco, Google flipped the switch on conversational search. Voice and video queries now run standard across
Google transforms search from typing queries to live conversations with AI. The feature exits experimental status and becomes standard across US mobile apps, forcing competitors to match conversational capabilities or risk appearing outdated.
Voice typing finally rivals keyboards—if you choose the right balance of privacy, speed, and cost. New AI-powered apps transform rough speech into polished text, but the local vs. cloud decision shapes everything else.
💡 TL;DR - The 30 Seconds Version
🎯 AI dictation apps now transform rough speech into polished text using Whisper models, making voice a credible primary input method for the first time.
🔐 The core choice: local processing keeps audio private but needs powerful hardware, while cloud processing delivers sub-second speed and voice editing commands.
💻 Mac users get the strongest local options like MacWhisper and BetterDictation, while cross-platform teams should consider Wispr Flow's cloud approach.
⚡ New apps infer punctuation automatically and support voice commands like "make this formal" or "turn into bullets" without touching the keyboard.
💰 One-time purchases like MacWhisper offer the best long-term value, while subscription services add enterprise features and consistent cloud performance.
🚀 The right setup can make users 3-5× faster for emails and reports, but success depends on matching privacy needs, platform mix, and editing workflow preferences.
Your keyboard isn’t dead, but voice now rivals it—if you pick the right mix of privacy, speed, and price.
Built-in dictation on Mac and Windows is free and getting better, yet many people still find it literal and fiddly. The new crop of apps layers accurate speech recognition with on-the-fly cleanup, formatting, and even voice-only editing. Much of the progress rides on the open-source Whisper speech model, now wrapped in polished apps with GPT-class text enhancements.
Local (on-device) dictation keeps audio on your computer, works offline, and avoids vendor storage. You trade some convenience and need capable hardware, especially for long sessions and larger models.
Cloud dictation delivers sub-second latency and “hands-free” editing that fixes tone and structure as you speak. You need a reliable connection, must accept server processing, and usually pay a subscription. Pick this if you work across Mac and Windows or need admin controls.
A quick mental model: if privacy and offline use are non-negotiable, go local. If ease, speed, and team features matter most, go cloud. Simple.
MacWhisper (one-time license). Runs Whisper entirely on your Mac, supports all model sizes, and adds system-wide dictation via a global hotkey (direct download). It doubles as a pro transcriber for audio/video files, subtitles, and speaker labels. You can plug in your own GPT/Claude keys to clean up drafts. Best value if you want private dictation plus real transcription tools. Caveat: the biggest models are resource-hungry; Apple Silicon flies, older Intel Macs crawl.
BetterDictation (one-time for Apple Silicon; optional $2/mo add-on). Push-to-talk simplicity with fully offline dictation on M-series Macs. The optional Pro layer removes fillers, tidies grammar, and formats lists. No Intel support and no cloud fallback, but it’s fast, tiny, and reliable for daily writing.
VoiceInk (open-source; paid prebuilt available). Local-only via whisper.cpp with smart context: it can read surrounding text to disambiguate jargon and names, and it supports personal dictionaries. Great for tinkerers and anyone who wants open code with a clean UI. Requires a modern macOS version.
Spokenly (free local mode; $7.99/mo for managed cloud). A generous “Local-Only” toggle gives unlimited on-device dictation for Mac and a matching iOS keyboard. Flip to cloud (yours or theirs) if you want bigger models and AI formatting without managing keys. Lightweight and friendly.
SuperWhisper (subscription or $249.99 lifetime). A power user’s kitchen sink: live formatting, translation, file transcription, and customizable “modes” that turn rough speech into polished emails, notes, or code comments. It can run locally or in the cloud and includes bring-your-own-API flexibility. There’s a learning curve, but the output can be near send-ready.
Who should choose macOS-local? Writers, researchers, therapists, lawyers, and anyone handling sensitive information. You’ll own the workflow, avoid monthly bills, and keep every word on your machine.
Wispr Flow (Mac, Windows, iPhone; free tier + Pro). “Just works” voice typing with sub-second latency and Command Mode to edit by voice (“replace the last sentence,” “make this formal,” “turn into bullets”). A privacy mode limits retention; enterprise plans add SOC 2/HIPAA, admin controls, and shared dictionaries. The free plan is small (weekly word cap), but enough to test. It’s the default choice for mixed-OS teams.
Willow Voice (Mac now; Windows in progress; free trial + subscription). Cloud-first dictation focused on context and personal style. It auto-formats, adapts to your tone over time, and offers SOC 2 and zero-retention options on higher tiers. If you want send-ready prose without tweaking, this is the “no knobs” pick.
Dictation Daddy (Windows/Mac; free + $99 lifetime deal). Budget cloud alternative with a true lifetime option. Cross-platform, continuous dictation, auto-punctuation, and simple prompts. Fewer enterprise assurances than Wispr/Willow and fewer advanced editing tricks, but the price and breadth are attractive.
Who should choose cloud? Teams with compliance needs, Windows users who want modern dictation without Dragon’s learning curve, and anyone who values voice-only editing and minimal setup.
Nuance Dragon Professional v16 (Windows). Deep command-and-control, custom vocabularies, and voice macros built for legal/medical workflows. It’s expensive and training-oriented, but if you want to drive your PC by voice and rely on domain terms, Dragon remains the most mature solution. If you mainly need natural dictation with AI cleanup, the newer cloud tools are easier.
iOS already has strong built-in dictation, but Spokenly, SuperWhisper, and Wispr’s iPhone app add multi-language Whisper accuracy or shared dictionaries with your desktop. Android support is sparse among the new wave; cross-platform cloud tools with web apps or extensions are your best bet today.
There’s no polished commercial suite yet. Power users get excellent results from whisper.cpp or Faster-Whisper combined with small scripts and a hotkey tool to “type” into any field. It’s perfectly usable—if you’re willing to assemble the pieces.
Speech-to-intent, not just speech-to-text. Good apps don’t make you say “comma” or “new paragraph.” They infer punctuation, structure your text, and can transform tone on the fly. In practice, that cuts edits dramatically.
Hands-free editing. Voice commands that feel natural—“shorten this,” “swap X for Y,” “turn into bullets”—let you compose and revise without touching the keyboard. Once you internalize a handful, you’ll move faster.
Per-app behavior. The smartest tools adapt to context. In Outlook they stay formal, in Slack they loosen up, in a code editor they respect jargon. Some even read nearby text to resolve names and acronyms correctly.
Enterprise posture. SOC 2 reports, HIPAA-ready options, BAAs, and zero-retention switches make cloud dictation viable in regulated settings. It’s not just a consumer story anymore.
When comparing costs, include LLM cleanup. Some local apps let you bring your own API key; that’s cheap for light polishing and gives you control over spend.
Whisper-based tools are now excellent in English and strong in many languages. Still, all ASR can mishear names, dialects, and noisy rooms. Hallucinations—adding text no one said—are rare in clean dictation but can surface with long silences or messy audio. In clinical, legal, or investigative work, review the transcript before it leaves your desk. That’s the rule, not the exception.
Latency depends on hardware and model size locally, and on network conditions in the cloud. On Apple Silicon, small and medium models feel real-time. In the cloud, speed is usually consistent until your connection isn’t.
Google transforms search from typing queries to live conversations with AI. The feature exits experimental status and becomes standard across US mobile apps, forcing competitors to match conversational capabilities or risk appearing outdated.
AI music jumped from novelty to seven-figure Spotify plays, but copyright lawsuits now shadow the best tools. This buyer's guide maps which platforms balance quality with legal safety—and why licensing talks may reshape everything.
Privacy-focused email promised liberation from Big Tech surveillance. Reality delivered Bridge daemon crashes, mobile search gaps, and calendar sync headaches. A year later, deadline-driven pragmatism wins over ideological purity.
Claude Code's new sub-agents let you build specialized AI assistants instead of one generalist trying to do everything. Each operates in its own context window, preventing pollution and delivering higher quality results for specific tasks.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
