Free tools Windows power users keep installed
One-click scans. No signup required.
King’s most concrete example of AI accelerating game development is not a system that creates finished Candy Crush levels. It is a set of tools that simulates different kinds of players, tests levels before release and recommends possible changes for human designers to assess. In an interview published on October 13, 2023, and updated June 18, 2025, King CTO Steve Collins described that approach alongside the company’s internal engine, cloud migration and experiments with coding assistants. His account is a snapshot of King’s work at that time, not a verified inventory of its technology in 2026. Read the GamesBeat interview with Steve Collins.
What was King trying to accelerate?
The production challenge was not simply making more levels. Candy Crush had grown from roughly 2,000 levels in 2016 to approximately 15,000 by 2023, according to Collins, while King released new content and episodes on a regular cadence he described as about every two weeks. Each level needed to work for a varied audience, fit into the game’s progression and present a suitable challenge.
That level-count increase cannot be attributed to AI alone. A decade of production experience, larger or better-organized teams, internal tools, live-service processes and platform improvements all contribute to a production pipeline. Collins’s account is most useful as a description of how AI-assisted testing fits into that wider system—not as evidence that AI autonomously produced Candy Crush’s levels.
How King’s AI players test a level
Collins said King began exploring AI around 2016, initially by building systems that could play its games. These agents are best understood as automated test players designed to approximate particular behaviors, not as a single universal player or a substitute for actual players.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minute#1 Best Overall
- CLASSIC BEGINNER GAME: Do you remember playing Candy Land when you were a kid. Introduce new generations to this sweet kids' board game
- RACE TO THE CASTLE: Players encounter all kinds of "delicious" surprises as they move their cute gingerbread man pawn around the path in a race to the castle
- NO READING REQUIRED TO PLAY: For kids ages 3 and up, Candy Land can be a great game for kids who haven't learned how to read yet
- GREAT GAME FOR LITTLE ONES: The Candy Land board game features colored cards, sweet destinations, and fun illustrations that kids love
Simulate different approaches
A useful test set needs more than an expert agent that finds an efficient solution. Collins described efforts to represent different skill levels, risk tolerances, competitive instincts and ways of approaching a level. The aim is to expose differences in how a design may play: a level that an expert agent clears easily could still frustrate a less skilled player.
This is simulation AI, distinct from generative AI that produces text, images, audio or code. It is also different from analytics systems that look for patterns in player data and optimization systems that recommend changes against chosen goals. King’s most developed examples in the interview concerned simulation, testing and recommendations, rather than fully automated content creation.
Test and recommend, then let designers decide
In Collins’s description, agents could test levels and give designers near-live feedback about how different simulated players might fare. The system could flag a level that appeared too easy or difficult, a progression curve that might become frustrating, a mechanic that seemed underused, or a level that played differently across simulated player types. It could also suggest adjustments.
Collins offered an example of a recommendation to make a level about 10% more difficult. That was an illustration, not a universal production rule or a disclosed benchmark. The interview does not specify King’s model architecture, training data, agent count, simulation fidelity, recommendation accuracy, cost per level or measured developer hours saved. Those omissions make it impossible to independently assess how reliably the system predicted real-player experience.
Why the designer remains responsible
A simulation can estimate behavior against a defined objective; it cannot establish by itself that a level is fun, fair or right for the game. Designers still need to judge whether a challenge feels satisfying rather than annoying, whether a change supports the intended emotional rhythm, and whether a metric improvement damages player trust or enjoyment.
This distinction matters because a model’s recommendations reflect the goals and data used to build and evaluate it. If a studio rewards difficulty, retention or progression without checking the wider player experience, a system can help optimize the wrong thing. Human review is not just a final quality-control step: it is where creative intent, context and player welfare can overrule a measurable suggestion.
Rank #2
- THE CLASSIC: Everyone knows the game of Candy Land. This classic edition features the original, charming artwork from the games' early days
- RACE TO HOME SWEET HOME: Kids encounter all kinds of delicious fun as they move their cute gingerbread pawn through Gum Drop Mountains, Lollypop Woods, the Molasses Swamp and more
- NO READING REQUIRED: Since it's for kids ages 3 and up, Candy Land is suitable for those who haven't learned how to read yet
- A FAMILY FAVORITE: Cherish your child's early years by making memories with Candy Land. It's called a Rite of Passage game because it has been enjoyed generation after generation
- SKILLS: All in the name of fun. Children will learn counting, turn taking, color recognition, and color matching
The production loop: telemetry, simulation and live updates
AI testing is most useful as one part of a feedback loop. In the workflow Collins described, simulated agents help assess content before launch; telemetry and A/B tests then show how real players respond after release. Those observations can inform later level adjustments and future content. Simulation offers controlled, repeatable tests; live behavior provides evidence about actual audiences. Neither replaces the other.
Telemetry can also help a studio look beyond an average player and examine patterns among new players, experts, people who leave at difficulty spikes, or people with different play styles and session habits. Collins saw potential for large language and multimodal models to help sift through large volumes of data and surface useful insights. But player data is not self-interpreting: correlation does not prove a cause, and an engagement pattern is not automatically a design goal worth pursuing. Segment analysis also needs care around privacy, bias, accessibility and differences in device or regional context.
Internal tools, Fiction and the burden of long-lived games
AI is only one way to reduce production friction. King’s live titles used an internal technology platform called Fiction, which Collins described as an engine designed for mobile casual games. He said the company supported games across iOS, Android, desktop, Facebook, Kindle and other devices. For a long-running live game, the underlying technology has to keep pace with changing operating systems, graphics APIs and hardware without disrupting players or the content pipeline.
A specialized engine can let a studio tune rendering, tools, deployment and compatibility around its own games. Collins also described work on platform transitions such as moving from OpenGL to Metal. Such engineering is less visible than AI-generated content, but keeping a live title working across a changing device landscape is part of sustaining its release cadence.
Collins said King had also explored Unity for some newer or different types of games. The interview does not identify all those projects, and it does not establish that Fiction is used by every King game. The choice between an internal engine and a commercial one depends on a studio’s portfolio, scale and constraints, not on a universal ranking:
| Approach | Where it can help | Costs and trade-offs |
|---|---|---|
| Proprietary engine | Deep tuning for a studio’s genre, platforms, rendering needs and content pipeline; direct control over tools and long-term compatibility. | High engineering and maintenance commitment; responsibility for platform changes; need to recruit and retain engine specialists; smaller third-party ecosystem. |
| Commercial engine | Faster initial setup, established editor and platform support, a broader talent pool, documentation, plugins and marketplace assets. | Licensing or subscription terms; dependence on vendor decisions; possible workflow or performance compromises; migration risk if the technology or business terms change. |
Unity’s current product and pricing details are published at Unity’s plans page, and its services can involve usage-based or monthly-active-user pricing described in its services pricing documentation. Unreal’s licensing depends on product and use; its current terms are on the Unreal Engine licensing page. Those pages can change, so a studio should check the terms that apply to its project rather than treat a pricing snapshot as permanent. King’s internal-engine choice was supported by its scale, longevity and related mobile-game needs; it is not automatically economical for a smaller team.
Rank #3
- 【Fun Family Bonding Game】: One player designs a colorful, tangled challenge while the other races to solve it! This brain-teasing board game strengthens relationships and creates lasting memories for kids and adults. Protected by U.S. Appearance Patent: D1101871.
- 【How to Play】: "Insert both ends of 10 vibrant ropes (plastic-tipped) into the board’s 22 holes, twisting into knots or crosses. Adjust difficulty with 3-10 ropes! Players move rope tips OVER others—never under—to empty holes. Remove untangled ropes and clear all to win—For detailed instructions, watch the video: How to Play – Rope Untangling Challenge."
- 【Multi-Level Difficulty for All Ages】: "From simple board games for kids (3-4 ropes) to complex puzzles for adults (10-rope mazes), this game grows with you! More knots = tougher challenges, perfect for fans of strategy-based board games!"
- 【A Fun Way to Keep Kids Off Screens】: "This hands-on logic puzzle game beats screen time! Kids build spatial awareness, sharpen focus, and learn problem-solving—like a tactile upgrade to traditional board games or video games!"
- 【The Perfect Gift for All Ages】: "Packed in a travel-ready case, this interactive family board game sparks friendly competition, boosts critical thinking, and unites everyone—ideal for playdates, holidays, or group games for kids and adults seeking screen-free fun!"
Cloud migration and workflow automation
Collins said King was moving its games from company data centers to cloud operations and described that transition as nearly complete at the time of the interview. Centralized infrastructure can make telemetry and analysis easier to access, support elastic capacity for experiments, standardize deployment and help distributed teams provision resources. It can also provide a foundation for machine-learning workflows.
Cloud migration is not a guaranteed cost reduction or speed boost. Usage-based bills, data-transfer charges, vendor dependence, security and compliance duties, latency and operational complexity all need to be managed. A cloud environment only accelerates work if the tooling, deployment practices and team workflows make it easier to test and ship safely.
Generative AI in engineering: promising, but not quantified
Collins said King was experimenting with large language models and tools such as GitHub Copilot, and described them as promising for coding. Potential uses include boilerplate, tests, documentation, explaining unfamiliar code, query writing, prototyping and internal tools. The interview did not report a measured productivity gain, so it does not support a claim that Copilot made King’s developers a specific percentage faster.
These tools shift some effort from drafting code to reviewing it. Suggestions can contain bugs, insecure patterns or APIs that do not exist; generated code can also create architectural inconsistency. Studios need policies for sensitive repositories and player data, provenance and licensing review, and ordinary security and code review. GitHub describes plan-specific allowances and billing in its Copilot billing documentation; the relevant commercial terms should be checked before adoption.
Budgeting for AI: offline, nearline and real time
The economics change sharply with when a model runs. An offline process can evaluate a batch of levels or summarize telemetry on a schedule. A nearline process can provide periodic recommendations. A real-time player-facing feature must answer quickly and repeatedly, potentially for a very large user base.
- Offline: Easier to schedule and budget, though large data or simulation workloads still consume infrastructure.
- Nearline: Useful for recommendations that do not need to appear during a player’s immediate interaction; freshness and processing cost need balancing.
- Real time: Adds latency requirements and recurring inference costs, as well as monitoring, moderation, redundancy and possibly human review.
Collins raised inference expense as a practical concern: a feature that is affordable in a prototype can become costly when multiplied across millions of player interactions. Cost estimates should include model compute, storage, retrieval, data transfer, monitoring and fallback capacity—not only a model’s headline inference price.
Rank #4
- Well-made Magnetic Board Game Great for Family Game Night. This magnetic game chess is a family bonding games to promote parent-child relationship through interation and create a relaxing and enjoyable family atmosphere, yet great for grandparents and kids to play together. Trust me, you will hear your family member laugh a lot when play this magnet game.
- Simple Instructions, Easy for Everyone. Each player holds the same number of stones and place them on the board, one at a time, taking turns with each player until get rid of the stones first as a winner. But beware, when stones jump together during one player’s turn, he or she collects them! Easy to learn and play both for kids and older people.
- Fun Board Game to Keep Kids Off Screens Also a Fun Puzzle Strategy Games That Fosters Exercises Your Child's Logical Thinking. If you are looking for a board games for kids 6-12 to keep them entertained at restaurants or on rainy days, this Hcusus magnet chess game is a good choice, also a great game for teen hangouts.
- Simple, Quick, But Challenging Game for 2-4 Player. There’s a lot of uncertainty in this magnet chess game, the most interesting thing is you think it is going to lose, but there’s still a chance to turn the table.
- Portable Board Game Ideal Gift for Family and Friends. Containd a chess game board with 40 stones and a storage bag. Easy to storage and take whether you're go. Great board game for family game night and party for adult and Kids entertained for hours.
Governance and common failure modes
AI-assisted production depends on trustworthy data, meaningful evaluation and clear accountability. A studio considering it should ask whether it has enough reliable telemetry, can define quality beyond revenue or retention, and can return results inside tools designers already use. It should also identify who approves a recommendation, what happens when a model or cloud service fails, and whether player or proprietary information may be sent to a selected provider.
- Agents can exploit game mechanics in ways real people do not, making simulation results look better than actual play.
- Averages can hide groups who find a level unfair, confusing or inaccessible.
- A model can repeatedly recommend safe, familiar patterns, reducing variety rather than improving it.
- Optimizing a measurable outcome can harm trust, enjoyment or player well-being.
- Generated code and assets require review for security, provenance, licensing and ownership issues.
- Usage, inference and data-transfer costs can be underestimated, especially for real-time features.
Collins said King had more than 50 people focused exclusively on AI tooling and capability, with more than 100 others working with AI across game teams, and that King brought in an AI and machine-learning team of about 45 through its acquisition of Peltarion. These are interview-era figures, not current staffing numbers or independently verified measures of results. They illustrate that a mature AI workflow involves specialists and organizational investment as well as models.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →What smaller studios can take from King’s example
King’s approach is rooted in a large live portfolio, extensive player telemetry, dedicated engineering and a recurring content pipeline. A smaller studio should not assume it needs a custom engine or a specialist simulation organization to benefit from AI. The more practical first step is to identify a repeated bottleneck, then test whether automation improves it without making quality harder to judge.
- Choose a repeatable task. Start with an activity that happens often, such as regression testing, test generation, data queries or checking level progression.
- Define success before using a model. Measure useful outcomes such as defects caught, review time or design problems found, alongside player-facing quality.
- Keep the process reviewable. Put recommendations where the responsible designer or engineer can inspect the evidence and reject a poor suggestion.
- Estimate the full operating cost. Include integration, inference, cloud use, review, maintenance and failure recovery.
- Protect data and creative ownership. Set rules for repository access, telemetry, model providers, generated content and documentation of provenance.
- Expand only when the workflow proves useful. A small, reliable tool is more valuable than an ambitious system that the team cannot validate or maintain.
Neural rendering: an outlook, not a current King capability
Collins discussed neural radiance fields, learned rendering and the possibility of describing a world that a neural system could generate and render. These were forward-looking ideas, not a demonstrated King production capability in the interview. Generating a convincing image or representation of a 3D place is not the same as building a complete playable world: a game also needs rules, state, agency, performance, testing, coherent content and clear rights to its assets.
The more defensible near-term lesson from Collins’s account is less dramatic than autonomous game creation. AI can become useful when it is integrated into a repeatable pipeline—testing, surfacing evidence and reducing routine work—while people remain accountable for the experience players actually receive.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Recommended Free Tools
