Which Agent
Free tool

Timeline

From the first wave of AI companions shutting down to the always-on agent wave. Dates are sourced. The future section is labelled forecast.

Oct 2025Jan 2026Apr 2026Jul 2026Oct 20265 Oct 2025 — Dot (New Computer) shuts down — the cautionary first wave of AI companions12 Jan 2026 — Claude Cowork research preview25 Feb 2026 — Perplexity Computer launches (Max)19 May 2026 — Gemini Spark announced at Google I/O ($100+ Ultra, US)2 Jun 2026 — Microsoft Scout introduced25 Jul 2026 — Spark opens to $19.99 Google AI Pro (US)11 Aug 2026 — Grok Bot beta15 Aug 2026 — Instinct invite-only launch (August)8 Sep 2026 — Muse launches (US, free tier)14 Sep 2026 — Siri AI beta ships with iOS 2716 Sep 2026 — Cowork + chat become “one Claude”; Instinct Concierge calls25 Sep 2026 — Microsoft Autopilot (ex-Scout) announced28 Sep 2026 — Instinct raises $1B at $10B; Manus launches Cue29 Sep 2026 — OpenAI Dots launch at DevDay
✓
Dot (New Computer) shuts down — the cautionary first wave of AI companions
✓
Claude Cowork research preview
✓
Perplexity Computer launches (Max)
✓
Gemini Spark announced at Google I/O ($100+ Ultra, US)
✓
Microsoft Scout introduced
◐
Spark opens to $19.99 Google AI Pro (US)
✓
Grok Bot beta
✓
Instinct invite-only launch (August)
✓
Muse launches (US, free tier)
✓
Siri AI beta ships with iOS 27
✓
Cowork + chat become “one Claude”; Instinct Concierge calls
✓
Microsoft Autopilot (ex-Scout) announced
✓
Instinct raises $1B at $10B; Manus launches Cue
✓
OpenAI Dots launch at DevDay

2026–2030: evolution phases ◇ forecast

Phases 1–2 and the start of Phase 3 were in a part of the brief that didn’t arrive. Only what we received is shown.
◇ forecast

Phase 3 — next (dates in the brief’s missing section)

  • Premium tiers sell concurrency, latency and quota — the “not rankable” metrics become the spec sheet.
  • Hybrid local + cloud becomes the default: sensitive/frequent work local, hardest reasoning in the cloud, a policy gate deciding what crosses.
◇ forecast

Phase 4 — 2028–30: gigawatt delivery decides abundance vs gatekeeping

  • Whether announced capacity is actually energised decides how many always-on agents can be served.

2028–30 scenarios ◇ forecast

Percentages are the brief author’s judgement “to force explicitness, not sourced forecasts”. Click a scenario to see what it would mean for choosing an agent (our opinion).

Pick a scenario above.

What to watch

  1. Weekly allowance and price-tier changes (e.g. Spark’s entry price fell from $100 to $20 in ~10 weeks).
  2. Gigawatts actually energised in 2027, not announced.
  3. How Colossus 2 is split between training, serving Bot fleets and renting to Anthropic.
  4. Prompt-injection bounty payouts and incident disclosures (Muse pays up to $130K).
  5. Per-memory view/delete/source controls (Dots is currently reset-or-nothing).
  6. Agent identity: whose email, phone, wallet or tenant ID do banks and governments accept?
  7. EU/UK launch dates.