3 May 2026: Replit Hands Out a Free AI Agent Day as the Oscars Bar Synthetic Actors

The AI afternoon split cleanly between vendors handing tools to builders and regulators tightening the screws. Replit has unlocked its agent free for a day, the Oscars have closed their stage to synthetic actors, and the EU’s foundation-model rulebook is ticking towards real enforcement.

Replit has handed its AI Agent to everyone, free, for 24 hours. The company is celebrating its tenth birthday with a global free pass to Replit Agent, the assistant that writes code, sets up databases, deploys apps and tackles work that until recently needed a developer in the room. The window opened on 2 May and runs until 9pm UK time on 3 May. Anyone with a Replit account can drive Agent without burning credits, and the company has paired the giveaway with a 100,000 dollar Buildathon for projects shipped during the same period.

For UK side-project builders and small business owners, this is a rare chance to test whether agentic coding can let a non-developer launch a workable internal tool, landing page or simple SaaS prototype. Sign in, describe what you want in plain English, and Agent does the scaffolding, hosting and database wiring. If it works in the next few hours, that is a usable answer about whether to keep paying from Tuesday. More on the Buildathon page.

The Oscars have shut the door on AI-generated actors and screenplays. The Academy of Motion Picture Arts and Sciences has tightened the rules for the 99th Oscars to require performances “credited in the film’s legal billing and demonstrably performed by humans with their consent.” Screenplays must also be human-authored, and the Academy can demand evidence on either count.

This is the first major industry guardrail against synthetic performers, after a year in which AI startups including the maker of Tilly Norwood pitched fully fabricated actors as the next stage of casting. BAFTA has not commented yet, but expect British awards bodies and unions including Equity and the Writers’ Guild of Great Britain to push for the same line here. TechCrunch has the full rule change.

MoonPay’s new debit card lets AI agents spend stablecoins anywhere Mastercard works. MoonPay launched the MoonAgents Card this week, a virtual Mastercard tied to a self-custodial wallet that allows AI agents to pay any Mastercard-accepting merchant in stablecoins. The card runs through MoonPay’s command-line interface, and the UK is one of two launch regions alongside Latin America. The US and EU will follow.

This is the first production payments rail purpose-built for autonomous AI spending. Custody never leaves the user, and approvals can be revoked. The implication for UK early adopters is concrete: an agent that books your travel or settles a software invoice can now do the last mile without you reaching for your wallet. Whether that is exciting or terrifying depends on how much you trust the agent you set loose. More on The Block.

OpenAI says GPT-5.5 is the fastest-ramping model it has ever shipped. The model, released on 24 April, is now growing API revenue at twice the pace of any previous OpenAI model. It hit 82.7 percent on Terminal-Bench 2.0 and 58.6 percent on SWE-Bench Pro, two tests that try to measure whether a model can actually finish a coding job end to end. OpenAI added that more than 85 percent of its own staff now use Codex weekly.

The wider read for UK dev shops is that enterprise spending on agentic coding has moved past the pilot phase. Pricing pressure on Claude Code, Cursor and the rest of the assistant pack will be real this quarter. If you are paying for one, the next renewal conversation is worth having from a position of choice. OpenAI’s product page.

EU AI Act enforcement powers switch on 2 August, with a 90-day countdown now running. General-purpose AI providers have been formally in scope since August 2025, but the European Commission gains its enforcement teeth on 2 August. From that date the EU can demand documentation, run independent evaluations, force a recall, and fine providers up to 3 percent of global turnover or 15 million euros. Lawmakers failed to agree the watered-down “Digital Omnibus” package last week and pushed talks into May, leaving the August deadline as the operating assumption.

UK firms shipping foundation models or AI features into the EU are in scope regardless of where they are headquartered. The FCA is sticking with its principles-based stance on AI here at home, but London companies with European customers should treat the next 90 days as a paperwork sprint. EU AI Act tracker.

Standard Intelligence raised 75 million dollars for video-trained computer-use models. The six-person startup closed its round led by Sequoia and Spark to build FDM-1, a model trained on 11 million hours of video showing humans driving graphical user interfaces. The pitch is that agents trained on real screen video, rather than static screenshots, can reliably click through software to do work like running vulnerability scans or operating CAD tools.

The point is not the startup itself but the bet investors are making on it. Anthropic Computer Use and OpenAI Operator both work from screen captures and accessibility trees. If video-trained agents prove better at long-horizon screen work, the agentic computer-use category gets reshuffled.

Here is everything else worth knowing from today’s AI news.

• AWS Bedrock now serves OpenAI models including GPT-5.5 in a limited preview, the first time AWS customers can call OpenAI inside Bedrock. Source
• Microsoft Agent 365 went generally available on 1 May, orchestrating Claude, GPT and Microsoft models depending on the task. Source
• Microsoft cut list pricing for Dragon Copilot, its clinical scribe, across the UK, Ireland and Western Europe from 1 May.
• Poolside open-sourced Laguna XS.2, a 33B mixture-of-experts coding model, under Apache 2.0. Source
• JuliaHub raised 65 million dollars and shipped Dyad 3.0, an AI-assisted modelling platform for vehicles and aircraft.
• Hightouch raised 150 million dollars at a 2.75 billion dollar valuation, for customer-data automation, led by Goldman and Bain Capital Ventures.
• Netomi raised 110 million dollars led by Accenture Ventures for agentic customer-experience tooling.

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s afternoon update on developments in artificial intelligence, published every weekday afternoon.

Featured image: Photo by Daniil Komov on Pexels.