Three releases that actually change how you work
The pace of tooling right now is genuinely hard to keep up with. But every so often, a stretch of days drops a few things that actually change how you work rather than just adding another tab to your browser. Three separate releases landed within days of each other: an AI agent that can join your Google Meet calls in real time, a GitHub Copilot feature that runs parallel coding agents simultaneously, and a free open-source tool that does everything a $89 screen-recording app charges for.
Pika Labs put a face on your AI agent
Pika Labs shipped PikaStream 1.0, a real-time video chat engine that works with AI agents including Claude and other assistants. The premise: you send a Google Meet invite to your agent, and it shows up to the call with a face, a voice, and full access to its memory and context, generating personalized video at 24 frames per second with roughly 1.5 seconds of end-to-end latency on a single GPU.
This isn't a demo or a concept. People are already using it to have agents book appointments and handle tasks live on calls. The agent maintains context throughout the conversation and can execute tasks during the call itself, not just respond to prompts afterward.
What it means: for client calls, sales demos, onboarding flows, and support calls, agents can now participate in real time rather than just responding after the fact.
What this actually changes about work calls
Right now, most people use AI by switching between a chat window and whatever they're working on. PikaStream collapses that gap. The agent is present in the conversation, not sitting in another tab.
For people running client calls, sales demos, or support flows, this is significant. You stop switching context. The agent is just there. What Pika has done is give the agent a persistent presence in the meeting rather than making it a background tool you consult during breaks.
Whether this becomes a standard feature across other platforms or stays a Pika-exclusive for a while is unclear. But the direction is obvious: agents are moving from text boxes into live environments.
The 4-step AI meeting assistant etiquette checklist
As AI note-takers and avatars transition from novelties to workplace defaults, unannounced bots are creating friction in client meetings and internal scrums. Follow these four etiquette rules before inviting an AI assistant to any call:
- Prior Notice & Transparency: Inform participants on the calendar invite that an AI note-taker will be present. Never surprise external clients or candidates with an unannounced recording bot.
- Explicit Recording Consent: State at the start of the call that audio is being transcribed and verified. If any participant requests that recording stop, dismiss the bot immediately.
- Zero “Ghost Bot” Attendance: Sending a bot to record a meeting you do not personally attend is widely perceived as disrespectful unless specifically cleared by the organizer in advance.
- Equitable Summary Distribution: Never hoard transcription insights. Circulate the generated action items, timestamps, and takeaways to all attendees to maintain collaborative trust.
Comparing top AI meeting assistants (Otter vs Fireflies vs Granola)
Choosing the right meeting assistant depends on whether you require a visible bot participant or silent local transcription:
| Tool | Joining Method | Key Strengths | Privacy & Local Storage |
|---|---|---|---|
| Granola | Local Audio Screen App (No bot joins call) | Customizable human-in-the-loop notes, elegant UI, zero bot disruption | High privacy; records system audio without external bot accounts |
| Otter.ai | Virtual Bot Participant (OtterPilot) | Live collaborative transcript, automated slide captures, Zoom/Meet sync | Cloud storage; visible to all attendees as meeting participant |
| Fireflies.ai | Virtual Bot Participant (Fred) | Deep CRM integrations (Salesforce, HubSpot), sentiment & topic trackers | Cloud workspace; enterprise encryption & admin governance |
GitHub Copilot's /fleet runs multiple coding agents at once
The /fleet command in GitHub Copilot CLI lets you break a large task into independent subtasks and run them in parallel. Each subagent gets an isolated context window. They share the filesystem but don't communicate with each other directly—an orchestrator handles coordination, dispatching independent work items simultaneously and synthesizing the final result.
You can also define custom agents by dropping config files into .github/agents/ in your repo, and reference a specific one in a prompt with @agent-name. Best use case: parallel tracks on the same codebase, like building an API layer, UI components, and configuration simultaneously. For linear, single-file work, the overhead isn't worth it.
Running agents in parallel is a different way of thinking
The instinct most people have with coding tools is to work one thing at a time: finish the API, then build the UI, then update the config. /fleet pushes you to think about your work as a dependency graph instead.
Tasks with no dependencies on each other can run simultaneously. That's not a new concept in software engineering, but having it available at the task-orchestration level inside a CLI tool is new for most day-to-day workflows.
The honest caveat: this pays off on genuinely large, multi-track work. If you're editing a single component or fixing a scoped bug, spinning up a fleet adds noise, not speed, and each subagent consumes its own request quota. The skill is knowing when to reach for it.
OpenScreen: a free, open-source alternative to paid screen recorders
OpenScreen is a free, open-source screen recording tool built as an alternative to paid options like Screen Studio. It includes auto cursor zoom that follows every click automatically with no manual keyframes, motion blur and smooth pan/zoom transitions, webcam overlay and annotation tools built into the app, and support for Windows, macOS, and Linux with no watermarks, account, or subscription required.
The tooling gap between a “professional-looking demo” and a “raw screen recording” has always cost something: either you paid for a commercial tool, hired someone to edit, or shipped recordings that looked rough around the edges. The auto-zoom that follows your cursor and clicks is the specific feature that makes demos look intentional rather than captured—the difference between a recording that looks like a tutorial and one that looks like a finished product.
Also on the radar
- Gemini 3.1 Flash-Lite (Google) — a faster, cheaper preview model with meaningfully quicker time-to-first-token than the previous Flash version, aimed at latency-sensitive applications.
- ChatGPT write access (OpenAI) — connectors for Notion, Linear, Box, and Dropbox gained write capabilities, letting ChatGPT create and update content directly inside them mid-session instead of only reading from them.
- Momentum acquisition (Salesforce) — Salesforce acquired Momentum to pull structured data from Zoom and Google Meet call audio into Agentforce, aiming to let agents act on what gets said in meetings, not just what gets typed into CRM fields.
As meeting bots evolve into full autonomous coworkers across engineering, sales, and operations, see our guide on AI digital teammates in the workplace. To route action items from your meeting notes into your CRM or task manager automatically, read the first 3 automations everyone should build or browse our complete AI Automation & Workflows topic hub.
And if you want custom automated workflows or high-converting landing pages built for your team, see our client services at Work With Me.