AI Research
What we are learning about putting AI agents to work, and keeping people in charge. Get new pieces by email
Start here
New to AI? Read these three first.
FoundationsWhat is AI?A clear, evidence-based answerA plain-English definition, the evidence on what today's systems can and can't do, and four questions to ask whenever someone says "AI."
FoundationsHow AI reasoning worksWhat thinking models can and can't doWhat "reasoning models" actually do when they think, what the evidence says they can now solve, and where they still break.
BuildingIs this an AI problem?A practical test for deciding when AI fits, and when it doesn'tWhere AI projects fail, what AI is good at, when a simple rule wins, and a five-question test for deciding whether a task really needs AI.
BuildingHow to evaluate an AI claimFive questions for benchmarks, studies, surveys and vendor promisesFive questions for judging AI benchmark scores, studies, survey headlines and vendor promises, with real cases of where AI claims go wrong.
FutureIs scaling enough to reach human-level AI?The case for bigger models, the case for new ideas, and what the evidence showsWill bigger AI models keep getting better, or does the next leap need new ideas? The strongest case on each side and what the evidence shows.
FutureHow soon will AI match human experts?The timelines debate, the strongest case on each side, and what the evidence can settleWhy experts put human-level AI anywhere from this year to mid-century, the strongest case for sooner and later, and the signals to watch.
MethodThe AI papers that shaped today's systemsA reader's index, from backpropagation to reasoning modelsSeventeen AI papers, from backpropagation to DeepSeek-R1, what each showed, its limits, and what citations and peer review can and can't tell you.
SafetyHow AI gets misusedScams, cyberattacks, influence and weapons risks, and what the evidence showsThe main ways people misuse AI, what the 2025–26 evidence shows about how much harm each one causes, and a simple question for judging the next scary headline.
LandscapeWhat are today's frontier AI models?The current lineup, how they compare, and why the strongest are harder to getA plain-English guide to the most capable AI models as of late October 2026: who released what, how independent scoreboards rank them, how fast the frontier is moving, and why the strongest are released in stages.
ApplicationsAI in marketingWhat the evidence shows about content, ads, research and searchWhere AI measurably helps marketers make content, ads and research, where it backfires, and how search is changing the ground underneath.
BuildingWhat is context engineering?Why the information you give AI matters as much as the promptWhy what you put in front of an AI model matters as much as how you ask, and what the evidence says about doing it well.
FoundationsHow good is AI at coding?What the benchmarks and field studies showWhat today's AI can actually do with code, why test scores and real-world speedups tell different stories, and how to judge AI-written software before you rely on it.
FoundationsA short history of AIBooms, winters and the ideas that finally workedSeventy years of AI in five chapters: the founding bet, two winters, the deep learning breakthrough and the scale era, and what that history tells you now.
BuildingWhy AI adoption stallsThe common mistakes and what the evidence says to do insteadWhy the headline AI failure rates disagree, the mistakes that stall adoption before, during and after rollout, and seven questions that catch them early.
BuildingIs this an AI problem?A practical test for deciding when AI fits, and when it doesn'tWhere AI projects fail, what AI is good at, when a simple rule wins, and a five-question test for deciding whether a task really needs AI.
DebatesWhat counts as AGI?Five competing definitions, the evidence, and why the choice mattersFive serious definitions of AGI, the strongest case for each, what recent evidence shows when you apply them, and why the choice changes the answer.
FutureHow AI forecasts are madeFive methods, their track records, and how to judge a predictionFive ways people forecast AI, how well each has done so far, what careful forecasters now expect, and five questions to judge any prediction.
MethodHow to read an AI research paperA practical method and what peer review can and can't tell youA three-pass method for reading AI papers, seven questions for where they mislead, and what the evidence says peer review can and can't tell you.
SafetyWhat is AI alignment?How labs steer AI systems, and where the evidence says it falls shortWhat alignment means, how labs steer AI models with human and AI feedback, and what recent experiments show about where those methods fall short.
LandscapeWho builds frontier AI?A map of the labs and playersWho trains the most capable AI models, how close the race really is, where the real gaps lie in compute and money, and where Canada fits.
ApplicationsAI in salesWhat the evidence shows it can and can't doWhere AI measurably lifts sales, where it backfires, and what buyers now expect from the humans who are left in the loop.
BuildingHow to prompt AI wellWhat the evidence says works (and what doesn't)What prompting actually is, which techniques hold up under testing, which popular tricks don't, and a simple method you can use this week.
FoundationsHow AI reasoning worksWhat thinking models can and can't doWhat "reasoning models" actually do when they think, what the evidence says they can now solve, and where they still break.
FoundationsWhat is AI?A clear, evidence-based answerA plain-English definition, the evidence on what today's systems can and can't do, and four questions to ask whenever someone says "AI."
Nothing matches. Choose All or try another word.
Build this with us
Manelink AI makes the Office, where AI agents take seats on your org chart and you see, direct and approve their work.
Meet the company