Twelve working pieces from the Lab. Each one is something you can play with, and each maps to work Omovera does for clients: agents that run sales and marketing, software shipped in days, and advice for leadership teams.
Omovera builds AI agents and software for businesses, runs them with a human team, and advises leadership teams on where to start.
A new enquiry lands on a website. Watch a sales agent research the company, score the lead, write the reply and book the meeting. A person approves the email before it goes.
6 hours
The agent replies 72× sooner, at any hour, for every lead.
Agent research, scoring, drafting, sending, CRM, scheduling, follow-ups.
Person approves the first email, takes the meeting, wins the deal.
One paragraph in the morning, a usable booking app by evening. Drag the clock to watch it take shape, then try the finished app.
"We are a three-therapist physiotherapy clinic. Patients phone to book and we lose calls at lunchtime. We want patients to pick a slot online, get a reminder the day before, and for us to see the day's list each morning."
Tap a time to book it.
Pick a role and adjust how much of each task an agent could take. See the hours handed over and what that capacity is worth each month.
The hours are time your people get back for calls, customers and judgment work. Whether that becomes growth or a smaller team is your decision.
One short brief from a bakery owner, turned into a full launch campaign: a landing page, three emails, social posts, ad lines and a video storyboard.
"Tidewater Bakehouse is launching breakfast boxes for offices: fresh pastries, fruit and coffee for 10 to 40 people, ordered by 4pm for next-morning delivery. We want office managers within 8 km to try a first box."
Warm pastries, cut fruit and proper coffee for 10 to 40 people. Order by 4pm, on your table by 8.30.
Introduces the box, one photo, the half-price first order and a two-click order link.
Social proof and the three most popular boxes. The number is filled from real orders once the campaign runs.
A gentle deadline, plus the standing-order option for teams that liked the first one.
Office breakfast boxes for 10 to 40, delivered by 8.30. First box half price for teams within 8 km.
Reel of the bake, the pack and the delivery, 12 seconds.
Sent to opted-in office managers only.
Search ad. Price is a placeholder for the owner to confirm.
Maps and local ad.
Display ad for retargeting.
Owner confirms price, offer and photos, then approves the first send. The agent then runs the schedule and reports results weekly.
A 20-person firm, shown by the work each team does in a week. Step through three waves of agents and watch which tasks move, and which stay with people.
Weekly hours across the firm: . The headcount stays at 20 in every step.
An expert-level science answer that cost 30 cents from the best AI model in January 2025 cost four hundredths of a cent by mid-2026. Since 2023 the price of a fixed level of AI performance has fallen about 13× a year.
US dollars per question to score 75% on a PhD-level science exam (GPQA Diamond). Log scale. The 2026 point is placed at mid-year; the exact date is not published.
Ten thousand such answers: $3,000 then, $4 now. A 725-fold fall in under 18 months.
2022 price is $20 per million tokens. 2026 prices are median list prices per million blended tokens: $0.31 budget, $1.96 mid-tier, $5.63 flagship. Even a budget model today far outperforms the 2022 one. Token counts are rough estimates.
Log scale, US dollars. Each line joins two measured points.
Yearly price fall for each technology, and how many times faster AI's fall is.
AI: 13× a year since 2023.
A fixed level of capability keeps getting cheaper. The newest flagship models do not: their median list price in November 2026 is more than double a year earlier, though still 85% below March 2023. The saving comes from using the cheapest model that is good enough for the job.
The 13× figure rests on five benchmarks and about three years of data. Its authors say it should not be read as exact, that other averaging choices give 43% to 58% a quarter, and that benchmarks are an imperfect stand-in for real work. The fall is fastest just after a capability appears, about 75× a year, slowing to about 4.7× two years on.