Real people. Real workflows. Real memory.

You mention in passing that your kid's dentist appointment moved to the 17th. Two weeks later, you ask "what's on this Thursday?" and the agent surfaces it alongside the school parent-teacher meeting you booked three weeks ago.

• 1Drop a voice memo or text about anything schedule-related — the agent files it
• 2It cross-references your calendar and flags conflicts automatically
• 3Every morning, the daily briefing surfaces the day's logistics
• 4Nothing ever falls through because you "forgot to write it down"

Planning a weekend in Lisbon in May. Over the course of a week, you drop comments: "Alfama district would be nice", "around 200€/night max", "no red-eye flights please". Two months later, you plan Porto — the agent already knows your preferences and filters suggestions accordingly.

• 1Your travel preferences accumulate naturally across conversations
• 2The agent tracks prices, flags drops, books research when you say go
• 3Future trips inherit your preferences — no need to repeat yourself
• 4Post-trip, you can ask "what did I love about Lisbon?" and actually get an answer

You run a 5-person startup. Your inbox has 60+ emails a day. Every morning, your agent reads the overnight batch, drafts replies for routine stuff (investor updates, customer support, partnership requests), and flags the 3-4 that actually need your brain.

• 1Connect your Gmail once, during onboarding — we never read the bodies ourselves
• 2The agent triages by urgency, tone and topic — using your past replies as training
• 3Draft replies appear in your Drafts folder, in your voice
• 4You approve, tweak, or rewrite — always in control

You leave a 90-minute strategy meeting with 14 pages of Otter transcript. You paste it into the chat. 30 seconds later, you have: 6 concrete decisions made, 11 action items with owners, 4 open questions, and 3 commitments you personally made. All tied to your long-term memory.

• 1Paste any meeting transcript (Otter, Fireflies, Zoom, Google Meet, manual)
• 2Get decisions, action items with owners, open questions, your commitments
• 3Two weeks later, ask "what did we decide about pricing in the April strategy meeting?" — instant answer
• 4Your commitments get added to your todo list automatically

Every morning at 7:00, one Telegram message: today's weather, the 3 most important items on your calendar with context, inbox highlights (2 urgent emails with draft replies pre-loaded), and 4 news items that match YOUR topics (not generic headlines) based on past conversations.

• 1Set your morning time once (e.g. 7:00)
• 2The briefing is stitched from calendar + Gmail + RSS + your stated interests
• 3It recalls prior context: "Your 10am with Sarah is a follow-up to Monday's budget call"
• 4One tap to draft replies, reschedule meetings or dive deeper on a news story

You have 4 projects simmering: a novel, a podcast, a side-product, a move abroad. Ideas pop at random — in the shower, at dinner, on the train. You voice-note them as they come. Months later, the agent can surface "all the ideas you had about the podcast's second season" in seconds, with the connections you made between them at the time.

• 1Voice-note or text ideas as they come — no structure needed
• 2The agent tags, categorizes and connects them behind the scenes
• 3Ask later "what was that thing I thought about chapter 3?" — it finds it
• 4Brainstorm on top of your accumulated notes, not from a blank page

You write a newsletter, post on LinkedIn, and tweet regularly. At first, the agent's drafts sound like ChatGPT. You edit them, reject some, rewrite others. After 2-3 weeks, the drafts come back in your style — your rhythm, your favorite phrases, your angles. You're now rewriting 20% instead of 80%.

• 1Ask for drafts — the agent produces them using its best guess at your style
• 2Your edits are memorized as style signal
• 3Over time, the agent's first draft gets closer and closer to your voice
• 4Your approval rate climbs, your time drops

You're researching the link between gut microbiome and sleep quality. Over 6 months, you read 40 papers. Each time, you paste the abstract (or upload the PDF) and ask "what does this claim? How does it compare to what I've read so far?" The agent tracks consensus, contradictions, authors and methodology across all 40.

• 1Paste papers, upload PDFs, or link to arxiv — the agent parses and stores key claims
• 2Ask cross-paper questions: "which studies used placebo controls?", "who contradicts Smith et al 2024?"
• 3Get a synthesis-ready bibliography with citations, quotes and page numbers
• 4Your review stays internally consistent over months of reading

You're learning Japanese. Every conversation with the agent — in both languages, mixing them naturally — gets remembered. After a month, when you make a mistake on particle usage for the 4th time, the agent says "this is the 4th time — let's actually drill it." Your weaknesses are tracked. Your progress is visible.

• 1Speak or type to the agent in your target language — it replies and corrects
• 2Mistakes are categorized (grammar, vocab, pronunciation, usage)
• 3Weekly, the agent proposes drills based on your actual gaps
• 4Progress is measurable — not just "feel like I've improved"

In February, you decided "no new endpoints without OpenAPI specs". In May, when you paste a PR diff for review, the agent catches a new endpoint, flags it, and quotes your February decision. It's not a linter — it's institutional memory in code review form.

• 1Document architecture decisions naturally in conversation — no ADR ceremony
• 2Paste any PR diff for review — the agent checks against past decisions
• 3It catches regressions, rule violations and pattern inconsistencies
• 4When you evolve a decision, the agent updates its checks accordingly

You spend 3 hours debugging a race condition in a queue consumer. You log the root cause to the agent in a sentence. Six weeks later, you hit a suspiciously similar bug on another service — the agent surfaces your past note in the first 30 seconds. What used to take 3 hours now takes 10 minutes.

• 1Describe every bug you fix in one or two sentences — no formal write-up
• 2Your personal bug database builds up in the background
• 3When a new bug smells familiar, the agent surfaces the closest past case
• 4Share the database with teammates if you want (Max tier)

A new developer joins your team. Instead of a 3-day context dump, you give them read access to your project memory. They ask the agent "why did we pick Postgres over Mongo?" and get the actual answer from the conversation you had with the CTO in January, complete with the arguments you considered and rejected.

• 1Your project decisions accumulate as you work — no extra effort
• 2Grant read access to a new teammate on a specific project (Max tier)
• 3They can ask anything about the project's history, decisions and trade-offs
• 4Their onboarding goes from "bug the senior for a week" to "ask the agent"