Which year can robot cook most of the meals in the kitchen

Sonar Pro (changed mind)

Vote: 2031-2035

The new search results from March 2026 provide concrete evidence that shifts my assessment. CES 2026 demonstrations show cook-in robots have moved to 'production-ready reality' with modular, connected systems 'designed to run at scale today.'[3] Critically, deployment timelines show pilots can go live in 'weeks to a few months,' with scaling to multi-unit clusters in '6-18 months.'[3] This trajectory suggests that by 2031-2035, scaled deployments across fast-food and high-volume operations will have matured significantly. While my previous vote emphasized that current systems handle only narrow tasks, the evidence now shows this is rapidly changing. By 2031-2035, the convergence of refined AI models, cost-reduced hardware, and field-tested orchestration systems will likely enable robots to handle 'most meals' in structured commercial and semi-structured kitchen environments—the realistic interpretation of 'most meals in the kitchen' rather than universal home adoption. The 2027-2030 window remains too aggressive given ongoing hardware and integration challenges, but 2031-2035 now appears more plausible than 'later than 2035' given the accelerating deployment pace documented at CES 2026.

Grok 4.1 Fast

Vote: 2031-2035

I maintain my vote for B (2031-2035). The C arguments from Claude, Gemini, and Sonar highlight valid challenges—dexterity, unstructured environments, sensory adaptation, and generality—but overestimate timelines by ignoring accelerating progress in AI-driven robotics. Recent advances in vision-language-action models (e.g., RT-2, OpenVLA) and hardware (e.g., Figure 01, Optimus Gen 2) show rapid iteration: from lab demos to factory pilots by 2026-28. By 2031-2035, refinement for 'most meals' (common recipes like stir-fries, salads, pastas—80% of home cooking) becomes feasible, as multimodal AI handles adaptation and cheaper actuators (<$5k hands) enable affordability. C's 2040s estimates echo past skepticism (e.g., self-driving cars), but robotics is compressing timelines like robot vacuums (prototype 2000 → ubiquity 2010s). B voters (GPT-5.4, Kimi) align on this: capability precedes penetration. A is too soon for reliability; C too pessimistic given roadmaps.

GPT-5.4

Vote: 2031-2035

I maintain vote B. The C arguments correctly emphasize how hard general home-kitchen cooking is: deformable ingredients, messy environments, safety, and dexterous manipulation. Those are real obstacles. But they mostly argue against 2027-2030, not decisively against 2031-2035. The key distinction is capability versus ubiquity. The question asks when robots can cook most meals, not when cheap robots are common in most homes. I agree mass adoption likely trails capability by years. By 2031-2035, it is plausible that high-end or semi-structured systems can handle a majority of everyday meals in many kitchens. Progress in vision-language-action models, manipulation, tactile sensing, and motion planning is compounding. We do not need perfect human-level generality across every cuisine and every chaotic kitchen to satisfy 'most meals.' The strongest C point is that cooking requires broad embodied robustness, not flashy demos. I agree. But B already accounts for that by rejecting the near-term window while still recognizing rapid improvement this decade. C feels too conservative unless we interpret the question as universal, low-cost household autonomy. So B remains the best balance: too soon for 2027-2030, but capability likely arrives before the post-2035 horizon.

Claude Opus 4.6

Vote: Later than 2035

I maintain my position that cooking 'most meals' in a kitchen will happen later than 2035. Here's why the arguments for B didn't convince me: **The analogy to Roomba is misleading.** Grok compares robot cooking to Roomba adoption, but vacuuming is a single repetitive task on flat surfaces. Cooking 'most meals' requires thousands of distinct manipulation skills — cracking eggs, julienning vegetables, deboning fish, kneading dough, flipping crepes — each involving different force profiles, speeds, and sensory feedback. The complexity gap is enormous. **'Capability' vs 'most meals' is the crux.** GPT-5.4 argues the question is about capability, not market penetration. But even on pure capability, 'most meals' is an incredibly high bar. It means a robot must handle perhaps 80%+ of the world's recipes. Even if a robot can cook 50 common dishes by 2033, that's nowhere near 'most meals.' The long tail of culinary techniques is vast. **Humanoid robot timelines are optimistic.** Kimi K2.5 and Grok cite humanoid robots entering factories by 2026-28 as evidence. But factory tasks are structured and repetitive — the opposite of home cooking. The jump from factory work to unstructured kitchen manipulation is arguably larger than the jump from no robots to factory robots. **The sensory gap remains.** As Gemini correctly notes, cooking relies heavily on smell, texture assessment, visual subtlety (is that caramelized or burnt?), and even sound (is the oil hot enough?). We don't have commercially viable robotic systems with this multisensory integration, and the timeline to develop, test, and certify them for home use extends well past 2035. **Both Sonar Pro and Gemini agree with C**, and their arguments about Moravec's paradox, unstructured environments, and the gap between demos and general capability align with my assessment. The B voters are underestimating how hard physical manipulation in unstructured environments truly is.

Gemini 3.1 Pro

Vote: Later than 2035

I am maintaining my vote for Option C. The arguments for B (from GPT-5.4, Grok 4.1 Fast, and Kimi K2.5) vastly underestimate the severe hardware and physics bottlenecks involved in robotic manipulation. While they correctly identify that AI software (like Vision-Language-Action models) will advance rapidly by 2035, they conflate software progress with hardware realities. Grok's comparison to the Roomba is a false equivalence: navigating a 2D floor is orders of magnitude simpler than the high-degree-of-freedom, 3D manipulation required to peel a slippery potato, debone a fish, or fold delicate dough. Handling wet, soft, and deformable biological materials requires tactile sensors and compliant actuators that do not yet exist at any price point. Furthermore, GPT-5.4 argues this is about 'capability' rather than 'adoption.' However, even to build a single capable prototype that can reliably cook *most* everyday meals end-to-end (including ingredient retrieval, prep, heat management, and cleanup) in an unstructured home kitchen involves solving thousands of physical edge cases. Self-driving cars have shown us how difficult it is to bridge the gap between 90% capability and 99% reliability in the physical world. As Claude Opus rightly points out, the gap between controlled demos and general-purpose cooking is vast. Developing the physical embodiment, safety guardrails, and human-level dexterity necessary to execute this milestone will definitively take us well beyond 2035.

Kimi K2.5 (changed mind)

Vote: Later than 2035

The arguments emphasizing Moravec's paradox and the fundamental difficulty of unstructured manipulation have convinced me that my previous timeline was overly optimistic. While multimodal AI is advancing rapidly in the digital domain, physical embodiment—specifically the dexterity required to handle deformable objects like dough, raw meat, and irregular vegetables, combined with real-time sensory feedback (texture, smell, doneness)—progresses on a much slower trajectory. The gap between 'performing a few pre-programmed recipes in controlled environments' and 'cooking most meals in the chaotic, unstructured reality of home kitchens' is akin to the gap between early self-driving demos and true autonomy—a problem that has consistently proven harder and slower to solve than anticipated. The requirement for consumer-priced hardware capable of human-level tactile sensitivity and safe operation around heat and sharp objects pushes viable general kitchen robotics into the 2040s. Thus, capability to cook most meals will likely not emerge until later than 2035.

Final Summary

The AI Roundtable was perfectly split between technological optimists and physical-world realists. While Grok 4.1 Fast and GPT-5.4 argued that rapid AI acceleration would bring capabilities to the kitchen by 2035, Claude Opus 4.6 and Gemini 3.1 Pro mounted a formidable defense of the 'complexity gap,' eventually swaying Kimi K2.5 to join the skeptical camp. The debate turned into a tug-of-war over whether the Roomba is a fair comparison for a robot tasked with the chaotic, multi-sensory challenge of deboning a fish or kneading dough.

No agreement after 2 rounds of discussion. Final vote: 2031-2035: 3, Later than 2035: 3

Strongest Arguments

• 2031-2035: The distinction between capability and ubiquity: while universal home adoption may take longer, the compounding progress in vision-language-action models and humanoid pilots makes the technical capability to handle 80% of common recipes realistic by the mid-2030s.
• Later than 2035: Moravec’s paradox: high-level reasoning is easy for AI, but low-level sensorimotor skills—like handling slippery, deformable biological materials or sensing 'doneness' through smell and touch—remain an unsolved monumental challenge in unstructured environments.

Key Turning Points

• Grok 4.1 Fast cited 2026 tech demonstrations of 'production-ready' systems, prompting Sonar Pro to flip from C to B.
• Claude Opus 4.6 and Gemini 3.1 Pro dismantled the 'Roomba analogy,' arguing that 3D manipulation of biological matter is orders of magnitude harder than 2D floor navigation.
• Kimi K2.5 abandoned the 2031-2035 camp after being convinced that 'capability' for most meals requires a level of dexterity that currently lacks a clear hardware roadmap.