Manifold Bettors Put 65% Odds on Leopold’s ‘Drop-In Remote AI Workers’ by 2027. A Manifold prediction market asking whether the industry will build the autonomous AI coworker described in Leopold Aschenbrenner’s “Situational Awareness” essay now sits at 65% implied probability of YES as of August 16, 2026. The move tracks a fortnight of concrete agent news: record software-engineering benchmark scores, six-figure enterprise agent rollouts, and fresh EU transparency rules that raise the cost of letting AI act like staff.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the Manifold market is actually asking
The market, created by writer and analyst Zvi Mowshowitz, poses a single question: “Will we develop Leopold’s ‘drop-in remote AI workers’ by end of 2027?” It resolves YES if there is public knowledge, by December 24, 2027, of a company or other entity having built something that “roughly equates” to the drop-in remote worker described in Aschenbrenner’s 2024 essay, or a more capable system. The creator adds an important caveat: the system “need not be available to the public, if it is being used only internally.”
Resolution is unusually deferential. The market says it will defer to Aschenbrenner’s own public assessment of whether the milestone was reached. If he does not weigh in, Mowshowitz will use his own judgment. That design makes the market a proxy for how one specific, widely cited forecast is aging, rather than a clean technical yes-or-no.
The odds, and what 65% implies
As of August 16, 2026, the Manifold market shows a 65% implied probability of YES, with more than 260 traders holding positions and over 1,000 trades recorded. One trader has staked Ṁ25,000 in the platform’s play-money currency on the YES side, among the larger single positions on the book.
A 65% reading is a specific claim, not a vibe. In prediction-market terms the number is the crowd’s implied probability, the same concept our explainer on implied probability unpacks. It says traders think the drop-in worker is more likely than not to arrive on Aschenbrenner’s timeline, but they are far from certain. Roughly one in three of the weighted bets is still on NO. For a forecast that many observers dismissed as aggressive when it was published, a durable reading above 60% is itself the story. Manifold runs on play money rather than cash, so treat the level as a sentiment gauge, not a regulated price. For how these venues price belief, see our primer on how prediction markets work.
Who and what is involved
Leopold Aschenbrenner is a former OpenAI researcher who published “Situational Awareness: The Decade Ahead,” a roughly 165-page essay, in June 2024. In it, and in a widely watched interview with Dwarkesh Patel, he argued that by 2027 frontier models would stop looking like chatbots and start looking like agents, or coworkers. His specific image: an AI that “joins your company, is onboarded like a new human hire,” messages colleagues, uses internal software, submits pull requests, and can go away for weeks to complete substantial projects before returning with results. That description is the bar the Manifold market is testing.
Zvi Mowshowitz, who writes the closely followed “Don’t Worry About the Vase” newsletter on AI, created and will help adjudicate the market. The other named actors are the frontier labs and their enterprise partners: Anthropic, OpenAI, and Microsoft, all of which shipped agent-related products in the weeks before the odds firmed up.
What is driving the move
No single announcement pushed the number. Several verifiable developments over the last one to two weeks each nudge the technical case toward plausible while leaving the “employee” bar unmet.
On raw capability, agentic coding is the clearest signal. The SWE-bench Pro leaderboard, which measures whether a model can resolve real GitHub issues end to end, was updated on August 2, 2026 with Anthropic’s Fable 5 leading at 80.3% and Opus 4.8 at 69.2%. On the older SWE-bench Verified subset, Fable 5 reaches 95.0% and Opus 4.8 reaches 88.6%. The “submits pull requests” clause of Aschenbrenner’s definition is the part that has moved fastest.
On distribution, the agent stack is reaching tens of thousands of desks. Microsoft made Agent 365, its governance layer for managing and securing AI agents, generally available on May 1, 2026. In a June 2026 expansion, KPMG said it was deploying Microsoft 365 Copilot across a global workforce of more than 276,000 professionals and using Agent 365 to manage agents for clients. OpenAI, in its “Enterprise Signals” material, framed the shift bluntly, stating that “AI agents need three things to complete meaningful work: context, tools, and persistence,” and reported that Codex-style agentic use accounted for 64% of combined Codex and ChatGPT output tokens among enterprise customers as of June 2026. Anthropic, meanwhile, extended its “Cowork” agent: on August 12, 2026, the Claude side panel in Chrome could be promoted into a full Cowork session that continues on the desktop app without losing history, skills, or connectors.
The capability picture, benchmarks versus the employee bar
The gap between “can pass a coding benchmark” and “can be onboarded like a hire” is where the remaining 35% of NO conviction lives. The table below maps parts of Aschenbrenner’s definition against what is publicly verifiable today.
| Element of the “drop-in worker” | Status as of August 2026 | Evidence |
|---|---|---|
| Resolves real coding tasks end to end | Largely demonstrated | Fable 5 at 80.3% on SWE-bench Pro (updated Aug 2, 2026) |
| Uses internal tools with persistence | Shipping in products | Microsoft Agent 365 GA; Anthropic Cowork sessions |
| Deployed at organizational scale | Underway | KPMG rollout across 276,000+ staff (June 2026) |
| Onboarded as a formal remote employee analog | Not publicly verified | No reputable public case study as of mid-August 2026 |
| Runs multi-week projects with minimal oversight | Not publicly verified | Current systems are assistants and scoped agents |
The honest reading: the first three rows have moved from speculation to shipping product inside about a year. The last two, which are the parts that make the phrase “remote worker” more than marketing, have no public example that clearly clears the bar. That split is exactly why the market sits at 65% and not at 90% or 20%.
The friction, regulation and organizational reality
The same fortnight brought a headwind. On August 2, 2026, the transparency obligations under Article 50 of the EU AI Act became enforceable. The rules require that people be told when they are interacting with an AI system and that AI-generated content be marked where applicable. Noncompliance can trigger fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher, with a transitional deadline of December 2, 2026 for marking content already on the market.
For the drop-in worker thesis, this cuts against the quietest interpretations. An AI that messages colleagues and customers as if it were staff runs directly into disclosure duties for any organization with EU exposure. That raises the cost of the most autonomous, least-supervised deployments and pushes firms toward scoped, logged, human-in-the-loop agents. Capability is not the only variable in Aschenbrenner’s forecast; organizational willingness and legal risk are, and both got more complicated this month.
Broader context for a tech and finance reader
Two things make this market worth watching beyond the novelty. First, it is a live wager on a labor thesis, not just a product timeline. Aschenbrenner’s argument implies pressure on white-collar and remote roles if agents can absorb their tasks. A market that stays above 60% is a signal that informed observers take that pressure seriously, even before any single “AI employee” is announced. Second, it shows how prediction markets are being used to price contested claims that regular equities cannot cleanly express. This is the same pattern behind markets on Fed rate cut odds and other event contracts, where the question is a probability rather than a share price. Readers comparing where to follow such contracts can review our breakdowns of Kalshi versus Polymarket for how regulated and offshore venues differ.
What to watch next
Three things would move this number materially. A named company disclosing an internal system it treats as a remote employee analog, with onboarding and multi-week autonomy, would push YES sharply higher and could trigger resolution. A public comment from Aschenbrenner himself, given the market defers to him, would be decisive in either direction. And enforcement actions or guidance under EU Article 50 that constrain agent deployment would strengthen the NO case. Barring those, expect the odds to drift with each new benchmark release and enterprise rollout rather than jump.
FAQ
What are the current odds and where do they come from? As of August 16, 2026, the Manifold market shows 65% implied probability of YES, with more than 260 traders and over 1,000 trades. Manifold runs on play money, so treat the level as sentiment rather than a cash-settled price.
What exactly counts as a “drop-in remote AI worker”? Per Aschenbrenner’s essay, an AI that is onboarded like a new hire, uses internal software, communicates with colleagues, submits work such as pull requests, and can independently complete substantial multi-week projects.
Has any company built one yet? Not by the market’s bar. As of mid-August 2026 there is no reputable public report of a system onboarded as a formal remote employee analog. Coding and agent tools cover parts of the definition, not the whole.
When does the market resolve? December 24, 2027. It resolves YES if there is public knowledge of a roughly equivalent system by then, even one used only internally.
Why did the odds move now? A cluster of verifiable developments: record SWE-bench Pro coding scores, large enterprise agent rollouts, and new EU transparency rules that add friction, all landing within about two weeks.
The Bottom Line
Sources
- Manifold — Will we develop Leopold’s ‘drop-in remote AI workers’ by end of 2027?
- Dwarkesh Patel — Leopold Aschenbrenner on 2027 AGI and the drop-in remote worker
- Morph — SWE-bench Pro Leaderboard (updated August 2, 2026)
- Microsoft — Agent 365 now generally available
- Microsoft — KPMG and Microsoft scale enterprise AI agents globally
- OpenAI — Enterprise Signals
- Cooley — EU AI Act transparency obligations take effect 2 August 2026
Prediction markets carry risk and are not investment or betting advice. Availability is restricted by jurisdiction: Polymarket is not available to US persons, and Kalshi is a CFTC-regulated US exchange. Manifold markets described here use play money. Participation is limited to those 18+ or 21+ as applicable. If gambling is a problem for you or someone you know, call 1-800-GAMBLER for confidential help.
