If you grow under glass and a Source.ag rep offers you autonomous climate control this winter, the headline number in the pitch is a vendor's own result from a demonstration greenhouse, published three years ago. The deal announcement says nothing about price, crop-loss liability or what happens to Optimal's current customers. Source.ag announced on October 9 that it has bought Optimal, the London startup whose AI runs heating, ventilation, screens, lighting, CO2 and irrigation through a grower's existing hardware. Terms were not disclosed. The first autonomous deployments at Source customers in Europe and North America are due to go live this winter.
Source is already in the room at most large growers. Its software runs in more than 500 greenhouses in 25 countries, supports €2.5bn of tomato, pepper and cucumber production a year, and works with 13 of the world's 25 largest high-tech growers. That reach is why the contract questions matter now: the same vendor that forecasts your harvest is about to ask for the right to act on it.
What Did Source.ag Actually Buy?
Source bought a control layer that writes setpoints to the climate computer you already own. Optimal says it connects through "secure cloud APIs" to the existing climate computer, needs no extra sensors, names Priva and Ridder as partners, and lets growers switch back to manual control at any time. It plans the growing environment days ahead using physics and machine-learning models, then re-optimises as the weather changes.
Source's own product watches the crop, forecasts harvests weeks out and automates irrigation. Adding Optimal closes the loop, so one vendor observes, predicts and acts. Optimal founder Dave Hunter, a former head of quantitative trading at Deutsche Bank, joins Source as VP Autonomous Control with his engineering team. Optimal was founded in London in 2016 and was backed by Founders Fund, Backed VC, 2050 and Acequia Capital.
The strongest case for the deal is labour. Both companies cite extreme weather, volatile energy prices and a shortage of skilled growers. A head grower who can delegate routine climate decisions to software can run more hectares, and Hunter has said Optimal users spend under 10 minutes a week on climate and irrigation control. That is a vendor claim, but the direction matches what other autonomous-control vendors report.
Where Does the 13% Yield Figure Come From?
The 13% yield and 27% energy figures match results Optimal reported in 2023 from its own demonstration greenhouse, measured against an unnamed benchmark. The acquisition release says Optimal delivered 13% more yield with 27% less energy than "the benchmark Dutch grower" in a commercial-scale Dutch greenhouse. It gives no crop, no dates and no description of the benchmark.
The same pair of numbers appeared in a HortiDaily profile on September 26, 2023, describing Optimal's commercial-scale demonstration greenhouse in Westdorpe, the Netherlands, growing tomatoes and cucumbers. That piece listed 13% higher yield, 14% higher Brix, 27% energy savings and 20% lower CO2 emissions against a "benchmark Dutch greenhouse," with no detail on how the benchmark was chosen. It mentioned a low-energy experiment that had been running for five months. It gave no methodology and no independent verification.
Optimal's own website still presents the result as its "Netherlands" case study, attributed to a head grower at Optimal. Its one named commercial customer story, Integral Farms in Ontario, reports fewer temperature and humidity spikes and better uniformity. It reports no yield or energy figure.
The 13% may hold up. It is still a single vendor-run comparison, three years old, published on the vendor's own site. A tomato house in Westland running a different cultivar on a different energy contract this winter may see something quite different.
What Happened the Last Time AI Ran a Greenhouse?
Controlled trials have shown AI beating experienced growers, under conditions that flatter the AI. In the second Autonomous Greenhouse Challenge, run by Wageningen University & Research and funded by Tencent, five AI-driven teams each controlled a 96 m² compartment of cherry tomatoes in Bleiswijk from December 16, 2019 to May 29, 2020. All five beat the reference compartment run by experienced Dutch growers on net profit. The best team made €6.86 per m² against the reference's €3.10.
The paper's caveats are the useful part. It covered one site, one cultivar and six compartments. Net profit depended on prices set for the challenge and excluded capital costs. No team achieved fully autonomous control; humans stayed in the decision loop.
Commercial results since then are smaller and still mostly vendor-reported:
- Koidra, writing about its own KoPilot system, reported 6.8% more eggplant yield and 4.84% less heating steam against a reference zone at DC Farms over 2025, and noted that steam and yield were measured over different periods.
- Blue Radix said in October 2024 that its Crop Controller was in use at nearly 100 greenhouses after a €5 million raise.
- At Art Farm, a South Korean pepper grower, Blue Radix's value report showed the system ran the greenhouse 80% of the time and manual climate-computer changes fell to 1.7 a month. The grower still watches closely when outside weather swings.
Single-digit yield gains and large time savings are the realistic range in the commercial record. A 13% yield gain alongside a 27% energy cut would sit at the top of it.
What Does the Release Leave Out?
The announcement answers what the combined product does and leaves out everything a buyer signs. Specifically, it does not mention:
- Price. Neither Source's release nor Optimal's site publishes one. Optimal offers a demo booking.
- Liability for crop loss. If the AI holds vents shut through a humid night and botrytis follows, the release does not say who pays. Check whether the liability cap in the contract is anywhere near the value of a crop.
- Optimal's existing customers. Optimal has run commercial operations in the Netherlands and Canada. The release does not say whether those growers keep their current contracts or must become Source customers.
- The Sollum partnership. In June 2025 Optimal signed a partnership with Sollum Technologies to drive Sollum's dynamic LED lighting alongside climate and irrigation. The acquisition announcement does not mention Sollum.
- Data. Closed-loop control means Source will hold your crop, climate and energy data and the control history. The release says nothing about ownership or export.
Agro Care CEO Kees van Veen, a Source customer, told HortiDaily that "We do not need more hype" while backing autonomous control. That is the right posture for the contract too.
What Should a Grower Demand Before Switching It On?
Treat autonomous climate control as an operations change with a written rollback plan, and price it against a trial on your own crop. The same terms apply to any "physical AI" product that actuates plant equipment, from HVAC in a data centre to dosing in a water plant.
This Week:
- Ask your Source account manager for the methodology behind the 13% and 27%: crop, cultivar, dates, the benchmark greenhouse and who measured it. If they cannot produce it, model your business case on the single-digit gains in the trial record.
- If you are an Optimal customer today, ask in writing whether your contract, pricing and support continue unchanged, and for how long.
- Have your head grower list the setpoints and alarm limits that must never be overridden without a person, such as minimum pipe temperature, maximum humidity and CO2 ceiling.
Before You Sign for Winter:
- Run a split trial: one compartment or zone on autonomous control, one on your grower, the same cultivar, at least one full season. Agree the yield, quality and gas-per-kilo metrics in advance.
- Get hard limits written into the contract, with the AI unable to cross them, plus a documented manual takeover that your night staff have practised.
- Negotiate liability for losses caused by control decisions, or at least a service credit tied to agreed crop-risk events such as condensation hours or temperature excursions.
- Secure the right to export your climate, crop and control-decision history in a usable format, and keep your climate computer configured so you can run it without Source.
- If you use Sollum lighting, ask both vendors whether lighting control stays in the loop after the acquisition.
The Bottom Line
Source.ag now sells growers the forecast and the hands that act on it, and it has a customer base big enough to make autonomous climate control mainstream in one winter. The control technology has a real record: WUR's challenge showed AI beating experienced growers, and three commercial vendors report AI running live greenhouses. The specific number in this deal is a 2023 result from Optimal's own greenhouse.
Industrial buyers are already working through who owns plant data on industrial AI platforms, and the rule for any vendor benchmark holds here: a number with no logs behind it is a starting point for a trial. The approval-gate design and the exit terms you would demand of a software agent apply with more force when the agent opens vents on a live crop.
Ask for the trial methodology and the liability clause before the first cold night your greenhouse runs on autopilot.
Continue Reading
- Human-in-the-Loop for AI Agents: Most Approval Gates Rubber-Stamp
- Vendor AI Benchmarks: Only Numbers With Logs Belong in the Deck
- AI Vendor Exit Clauses: Standard Terms Give 0 to 90 Days to Get Out
- Schneider Buys PTC for $22.6B With an Openness Pledge but No Terms
- Schneider's $3.1B Cognite Deal: Industrial AI's Data War
- OpenAI's Agent Kill Switch Failed for 2.5 Hours After a DNS Escape