The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →King’s most consequential AI use in Candy Crush, as described by then-CTO Steve Collins in a 2023 interview, was not having a model invent finished levels. It was using AI players to test levels, estimate how different kinds of players might experience them, and suggest changes for human designers to judge. That distinction matters: the production gains came from fitting simulation, telemetry, internal tools and infrastructure into an existing live-game workflow—not from handing game design to an autonomous system.
What was King trying to accelerate?
By Collins’s account, Candy Crush had grown from roughly 2,000 levels in 2016 to approximately 15,000 by 2023, while King released new content in drops and episodes on a cadence of about every two weeks. The challenge was not just to make more levels. Each needed to be playable, appropriately difficult, coherent in progression and engaging for players with different skills and habits.
The increase in level count cannot be credited to AI alone. It coincided with years of production learning, larger and more organized teams, improved tools, live-service processes, player demand and infrastructure changes. In an interview published October 13, 2023 and updated June 18, 2025, Collins described AI as one part of that system, not a substitute for it. Read the GamesBeat interview.
How did King use AI to test levels?
AI players simulate different behaviors
Collins said King began exploring AI around 2016, initially by building systems that could play its games. These agents were intended to represent different kinds of players rather than a single ideal solver: skilled and less-skilled, competitive and noncompetitive, risk-taking and cautious, and people who approach a puzzle in different ways.
#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
That variety is useful because an expert agent clearing a level does not prove that the level is fair or enjoyable for a newer player. A set of behavioral profiles can expose different failure points and indicate how the same design may feel across player segments. The agents remain approximations; the interview did not disclose their architectures, training data, fidelity, or independent validation.
Testing and recommendations support designers
In the workflow Collins described, simulated players provide rapid feedback while designers are developing content. Systems can help estimate difficulty, identify progression or balance issues, and recommend adjustments. He gave an example of a recommendation that a level might need to become approximately 10% more difficult. That was an illustration, not a universal rule or a published production benchmark.
Potentially useful signals include a level being too easy or hard for a target segment, a difficulty curve producing a frustrating spike, or a mechanic going largely unused. A model can also help flag when one player profile has a markedly different experience from another. The interview did not publish measured accuracy, cost per level, hours saved, A/B-test results, or evidence that any specific recommendation improved outcomes.
Why designers remain accountable
Simulation can answer bounded questions about behavior; it cannot decide by itself whether the answer serves the game. Designers still need to judge whether difficulty feels challenging rather than irritating, whether a level belongs in its intended place in the emotional arc, and whether a metric improvement compromises fairness, surprise or player trust.
Free tools Windows power users keep installed
One-click scans. No signup required.
This is the central distinction between optimization and design. A system can recommend a change against a target such as completion or progression, but the target may be incomplete or wrong. Human review is therefore not just a final sign-off: designers define goals, interpret recommendations and reject changes that make the experience worse in ways the measurements fail to capture.
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-assisted testing makes most sense as part of a loop rather than a standalone feature. Historical and live telemetry can reveal where players struggle or disengage; agents can test proposed levels against selected behavioral profiles; recommendation systems can surface candidate changes; designers can approve or reject them; and experiments with real players can provide evidence about the shipped change.
- Collect signals: Record relevant, privacy-governed gameplay events and define the player segments the team needs to understand.
- Test proposed content: Run appropriate simulated agents against levels and inspect where their experiences differ.
- Review suggestions: Let designers assess proposed changes against intended difficulty, pacing and game identity.
- Measure in production: Use controlled experiments and live telemetry to assess real-player response, rather than treating simulation as proof of fun.
- Feed learning back: Use results to refine the tools, evaluation criteria and future content decisions.
Even large datasets do not interpret themselves. Correlation does not establish why a player stopped playing, and the average player can hide important groups: new players, experts, people who hit a particular difficulty wall, or people with different session patterns, devices and accessibility needs. Optimizing engagement without considering well-being and trust can produce the wrong outcome.
Why the engine and infrastructure matter
Fiction and cross-platform maintenance
Collins said King’s live titles used an internal technology platform called Fiction, an engine designed for mobile casual games. Specialization can make sense when a company has related games, shared production needs and a long-lived portfolio to maintain. An internal engine can be tuned to a studio’s rendering, content tools, deployment pipeline and device footprint.
King’s games had to keep working across a broad mix of iOS, Android, desktop, Facebook, Kindle and other devices as operating systems, graphics APIs and hardware changed. Collins discussed platform work including the move from OpenGL to Metal. That kind of ongoing compatibility work is part of development speed: a studio that can update a mature game without disrupting players preserves the capacity to ship content and improve operations.
Collins also said King explored Unity for some newer or different kinds of games. The interview does not identify a complete list or establish that Fiction is used by every King game. His comments describe a portfolio-specific choice, not a claim that one engine is categorically superior.
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 is an enabler, not a guarantee of savings
At the time of the interview, Collins said King’s move from company data centers to cloud operations was nearly complete. Centralized infrastructure can make telemetry, experimentation, deployment and machine-learning workflows easier to coordinate, and cloud capacity can be provisioned for changing workloads. None of that makes cloud migration automatically faster or cheaper: usage-based bills, data transfer, vendor dependence, security and compliance, latency, and operational complexity all need management.
When does an internal engine make sense?
Neither a proprietary engine nor a commercial one is the right answer for every team. The decision depends on whether the engine itself is a strategic bottleneck, whether the studio can sustain a platform team, and how much value specialization creates over the life of its games.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11| Approach | Potential advantages | Costs and risks | Best fit |
|---|---|---|---|
| Proprietary engine | Deep specialization; control over tools, rendering and deployment; reduced dependence on a vendor’s roadmap; potential efficiency at very large scale. | High initial and continuing engineering cost; responsibility for platform changes and compatibility; specialized hiring; smaller third-party ecosystem; modernization risk. | Studios with a durable portfolio, shared technical needs and resources to maintain engine, graphics, build and platform expertise. |
| Commercial engine | Faster initial setup; established editor and platform support; broad talent pool, plugins and documentation. | Licensing or subscription exposure; vendor dependence; potential workflow or performance compromises; migration risk if terms or strategy change. | Teams that value ecosystem and time to first release more than narrow specialization, subject to the engine’s current terms. |
For current terms, consult the vendors directly rather than relying on historical interview context: Unity’s plans and pricing and Unreal Engine licensing. Pricing, eligibility and license conditions can change, so these links are not a substitute for checking terms applicable to a particular project.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Where generative AI fits—and where the evidence stops
Collins described King as experimenting with large language models and coding tools such as GitHub Copilot, and as still learning how they fit into production. Potential uses include drafting boilerplate, explaining code, creating tests, documenting systems, writing queries, prototyping and exploring unfamiliar APIs. The interview did not quantify a productivity gain, so it does not establish that these tools made King’s developers a specific amount faster.
Generated code can be wrong, insecure, inconsistent with an architecture or based on an API that does not exist. It can also create licensing, provenance and confidentiality questions. Review work may shift from writing to verifying rather than disappear. Teams should apply their repository and data policies, keep human code review and security checks in place, and measure whether a tool reduces total work for the tasks they actually do. GitHub documents plan-specific Copilot billing and usage at its billing reference.
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.
Language models may also help analysts sift through large volumes of telemetry and surface patterns for investigation. They do not turn player data into self-evident truth: a summary is a lead to test, not a causal explanation or a mandate to optimize the behavior it describes.
Recommended Free Tools
Budgeting for AI: offline is different from real time
Inference cost becomes material when multiplied across a large audience. A prototype that seems inexpensive can incur substantial expense in model compute, storage, retrieval, transfer, monitoring, moderation and human review when exposed to millions of players. The relevant question is not just whether a model works, but what each useful decision or interaction costs at expected scale.
| Mode | Typical use | Cost and operational profile |
|---|---|---|
| Offline | Batch level testing or retrospective analysis. | Usually easier to schedule, budget and rerun than player-facing inference. |
| Nearline | Periodic recommendations or content evaluation. | Can balance freshness against cost and latency by processing on a schedule. |
| Real time | AI responding directly during a player interaction. | Requires low latency and ongoing inference capacity; budgets must account for usage spikes and service reliability. |
A team should estimate cost per level tested or interaction served, include failure and fallback behavior, and monitor actual usage. Caching, batching or smaller models may help for appropriate tasks, but trade-offs in freshness, capability and reliability should be evaluated rather than assumed.
What a smaller studio can take from King’s example
King’s approach reflects a large live portfolio, extensive telemetry, specialized engineering and frequent content production. A smaller team should not assume it needs a comparable AI organization or a custom engine. It can start by identifying a repeated production bottleneck and choosing the least complex tool that can address it.
- Check the data: Are telemetry events reliable, useful and collected with appropriate privacy controls?
- Define “better”: Can the team evaluate quality beyond one metric such as retention or revenue?
- Fit the workflow: Will recommendations appear where designers and engineers already work?
- Choose the right timing: Is the task suitable for offline, nearline or real-time processing?
- Assign ownership: Who reviews outputs, catches harmful optimizations and approves production changes?
- Plan for failure: Can development continue if a model, vendor API or cloud region is unavailable?
- Measure the whole cost: Include integration, review, inference, transfer, monitoring and maintenance, not only model access.
Common traps include agents that exploit mechanics unlike humans, repetitive content learned from safe patterns, a single average-player target that harms minority segments, and generated code entering production without review. In each case, the failure is partly one of evaluation and governance, not simply model capability.
What remains speculative
Collins discussed neural radiance fields, learned rendering and the possibility of describing a world for a neural system to generate and render. These were forward-looking ideas in the interview, not evidence of a King production capability. Rendering convincing imagery is not the same as building a coherent, playable world: games also need rules, state, player agency, performance, testing, ownership and deliberate content design.
The better-supported near-term picture is less spectacular and more useful: AI tools embedded in the daily work of designers, analysts, artists and engineers, with people setting goals and checking results. King’s account shows how simulation, internal tools, telemetry, infrastructure and human judgment can reinforce one another, while leaving clear limits on what the models themselves were shown to achieve.
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.
