Founding PM @skyfallai. Working on autonomous business systems. CS grad, hands-on builder. Previously early product @getlokalapp. Product, AI & experiments.
Hear directly from the Skyfall India team on the ‘aha’ moment when they realized Steve, the AI CEO, had made a real breakthrough while running the 3 acquired businesses.
Press conference at @WebSummit today with our CEO Sam @spisallyouneed on Steve, the World’s First AI CEO.
The core research behind our harness Steve lies on World Models that let Steve simulate business decisions before acting, paired with Recursive Self-Improvement that rewrites its own skills and code when forecasts miss to improve further.
Steve is already running three businesses we acquired over the last 4 months. Check live results here: https://t.co/GET69Mgin9
We’re now opening the Steve Startups Program of $2M to acquire up to 5 more businesses for Steve to run end-to-end.
🔗 to apply, waitlist and blog for technical details in comments.
That question is central to our product work. Owners define the goals, permissions and which judgments need approval. You should be able to set the boundary around the work you're ready to delegate and keep other decisions with you.
You can read more in our blog here: https://t.co/TSRZovsiEa
@robiartec That's the question we need to keep testing. Different businesses bring different constraints.
If you'd like us to consider your business for Steve, the application is here: https://t.co/0T8EcSzwEi
For me as Steve's PM, the appeal is making operating expertise more accessible to small businesses. Owners should be able to spend more time on the complex problems and what they want to build, with Steve taking on more of the daily work.
If you'd like us to consider your business for Steve, the application is here: https://t.co/jBhYLt5hUO
@heyDhavall@skyfallai Scaling this is a product challenge I'm excited about: each business brings its own customers, operating context and owner priorities. If you'd like us to consider your business for Steve, the application is here.
https://t.co/0T8EcSzwEi
@aakashgupta The ownership and permissions point in the replies is a good one. A task needs both a responsible agent and the access to finish it. That's one of the most interesting problems we had to solve so that tasks progress without any deadlocks between agents.
@garlotic For anyone asking where to apply, there's the form link below. Tell us what you've built and which work you'd like to hand over. We're considering investment or acquisition through the Steve Startups Program.
https://t.co/0T8EcSzwEi
@ethanorbe One detail worth calling out: owners set Steve's permissions and approval boundaries. Production releases and spending above agreed limits are gated. The blog goes into how those controls fit into the operating loop.
https://t.co/NIrOl4HX0O
@n0tisma Working on Steve, the question we keep coming back to is what should happen after a task finishes. The result needs to inform the next decision across product, growth and support. That's the operating loop we're building around.
Agreed, you don’t need to acquire a SaaS company to start testing this. We started with software businesses because code, billing and support leave records we can inspect. That lets us follow decisions across functions: a support issue becomes a product change, which then affects conversion or retention. A local shop could be a useful starting point too.
https://t.co/TSRZovsQtI
Thanks Rohan! Making improvements testable has been a big part of building the harness. Proposed changes to prompts, context or models run against sealed historical snapshots using only information available at that point. We compare plans and forecast calibration with the baseline before accepting a change. Each business keeps its own model and evidence separate.
https://t.co/TSRZovsQtI
Amazing job by the team to get us here. Everyone has worked so hard and I am so grateful to be a part of such an ambitious project.
World models have been mostly theoretical and we are trying to bring it to the real world with SKY-001 last week and now Steve.
We are currently running 3 businesses on Steve and have plans to scale it 10x over the 12 months. If you want to automate your business with Steve please apply to the Startups Program.
@nattcornelis@zentalksai Thanks Natt! We’ve learned a lot from the first three, and we’re ready to build on that with five more. Here’s the program: https://t.co/0T8EcSzwEi
That’s exactly where autonomy needs a boundary. We’d ideally keep payroll commitments behind human approval initially and limit what Steve can spend independently. The owner remains responsible for those commitments. Steve can earn more trust and autonomy over time by making decisions governed by the policy engine and learning.
You can read more about our permissions model: https://t.co/TSRZovsQtI
@zentalksai Thanks Zen! We’ve put a lot into checking whether a proposed improvement actually helps. Changes to prompts, context or models run through historical replays and comparison with the baseline before acceptance: https://t.co/NIrOl4HX0O