{"course":{"slug":"ai-careers","title":"Working in AI","url":"https://addaly.com/learn/ai-careers","level":"beginner","track":"work"},"lesson":{"slug":"choosing-a-lane","title":"Choosing a lane you can defend, and changing it later","url":"https://addaly.com/learn/ai-careers/choosing-a-lane","position":8,"of":85,"minutes":8,"answer":"Choose the lane where you can produce evidence fastest, because lanes are adjacent and moving between them costs one project, not one degree.","keyIdea":"Choose the lane where you can produce evidence fastest, because lanes are adjacent and moving between them costs one project, not one degree."},"body":{"format":"markdown","text":"## Four lanes, not twenty\n\nBeginners try to choose between fifteen job titles. There are four practical lanes for someone starting, and the titles sit inside them.\n\n**Analytics.** SQL, spreadsheets, dashboards, and the ability to explain a number. Widest door, fastest to evidence, lowest ceiling early and a perfectly good ceiling later.\n\n**Data engineering.** Pipelines, schemas, orchestration, reliability. More jobs than any other lane, more software engineering, longer runway to a first job unless you already program.\n\n**AI application engineering.** Building products on top of models someone else trained: retrieval, APIs, evaluation, deployment. Newest lane, fewest settled conventions, most reachable if you can already write code and ship a web service.\n\n**AI-adjacent work that is not engineering.** Product, operations, evaluation and annotation leadership, technical writing, solutions and customer-facing work, teaching. Real jobs, frequently overlooked, and often the best fit for someone with strong domain experience and no programming background.\n\n## The three questions that pick one\n\n**Where do you already have context?** Ten years in pharmacy, insurance claims, school administration or freight is an asset that a computer science graduate cannot acquire quickly. It points at a domain, and the domain often points at a lane: claims work points at analytics, freight at data engineering.\n\n**What can you evidence in eight weeks?** Not learn: evidence. A finished, deployed, written-up thing.\n\n**What is actually hired where you are, or where you can legally work remotely?** You measured this in the last lesson. If your city has forty analytics openings and two ML ones, that is not a conspiracy against your interests; it is the shape of your market.\n\n## Honest time-to-first-evidence\n\nPart-time, around six hours a week, starting from ordinary computer literacy:\n\n- **Analytics:** eight to twelve weeks to a defensible project. SQL is learnable to working level in weeks.\n- **AI application engineering:** three to four months if you already program, eight or more if you do not.\n- **Data engineering:** four to six months, because it needs programming plus systems plus at least one orchestration tool.\n- **Research:** years, and a different path entirely.\n\nThese are ranges from watching many people do it, not guarantees. Some move faster. Almost nobody moves faster than half of these.\n\n## The adjacency map\n\nThe reason this choice is less frightening than it feels: the lanes touch.\n\n- Support, operations or finance → analyst. Very common.\n- Analyst → analytics engineer → data engineer. Very common.\n- Data engineer → ML engineer. Common, and much easier than the reverse.\n- Backend engineer → AI application engineer. Short hop.\n- Analyst → product. Common at three to five years.\n- Anything → research. Rare without a formal route.\n\nNotice what is missing: a direct jump from courses to ML engineering. It happens, but it is the exception people write posts about, which is exactly why it looks frequent.\n\n## What transfers when you move, and what does not\n\nTransfers: the domain you learned, the habit of finishing, your written record of work, SQL, debugging, and the thirty people who have seen you work.\n\nDoes not transfer: certificates, course completions, and tool-specific fluency in a tool the next team does not use.\n\nThe practical consequence is that the switching cost is a project and a job change (six to eighteen months) not a restart. People treat a lane choice as though it were a degree subject, then delay for months rather than pick. The delay costs more than a wrong pick would have.\n\n## The failure mode that wastes a year\n\nStudying three lanes at thirty per cent each. It feels prudent and it is the single most common way a promising year produces nothing hireable, because employers hire on depth in one thing and nobody is hired for breadth at zero years.\n\nA workable rule: **one lane for six months.** Then reassess on evidence, not mood: did you finish something, did anyone use it, did anyone reply to you about it. Boredom in week three is not evidence.\n\n## What to do if you cannot get any job at all\n\nThen the lane is chosen for you, and that is fine. Take the job that exists, then use the bridge-role method in the next module to move toward data from inside. Choosing a lane is a luxury of having options, and building options is the actual first task.\n\n## The limitation\n\nNone of this survives contact with an unusual opportunity. If someone offers you paid work in a lane you did not choose, take it. Paid work in the wrong lane beats unpaid preparation in the right one, every time, because it converts you from a candidate into someone with a reference."},"check":{"question":"Rahel has eleven years in hospital administration, no programming, six hours a week, and wants out of admin within a year. She is deciding between teaching herself deep learning and teaching herself SQL and reporting. Which reasoning is soundest?","options":[{"text":"Deep learning, because AI roles pay more and hospital data is exactly what modern models are built for.","correct":false,"tempting":"Chooses by destination salary rather than by time-to-evidence, and ignores that the deep learning route from zero takes years at six hours a week."},{"text":"Neither yet: she should first complete a broad foundation covering statistics, programming, machine learning and data engineering.","correct":false,"tempting":"The breadth-first plan, which feels responsible and reliably produces a year with nothing finished and nothing hireable."},{"text":"SQL, but she should leave hospital work behind entirely, since her admin experience will not count in a technical role.","correct":false,"tempting":"Discards the domain knowledge that is the least replaceable thing she owns."},{"text":"SQL and reporting, because it is the lane where she can produce a finished, defensible piece of work within about three months, and her eleven years of hospital context makes her a stronger candidate for health analytics than a generalist with better maths.","correct":true}]},"prev":{"slug":"degrees-and-doors","title":"Who actually needs a degree","url":"https://addaly.com/learn/ai-careers/degrees-and-doors"},"next":{"slug":"the-ladder-and-the-levels","title":"Levels, titles, and why \"senior\" means nothing on its own","url":"https://addaly.com/learn/ai-careers/the-ladder-and-the-levels"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Choosing a lane you can defend, and changing it later', in Working in AI. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-careers/choosing-a-lane (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Choosing a lane you can defend, and changing it later","courseTitle":"Working in AI","module":{"index":1,"slug":"the-terrain","title":"The terrain: who hires, for what, and where you fit"},"answer":null,"keyIdea":"Choose the lane where you can produce evidence fastest, because lanes are adjacent and moving between them costs one project, not one degree.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"four-lanes-not-twenty","runs":[{"text":"Four lanes, not twenty"}]},{"t":"p","runs":[{"text":"Beginners try to choose between fifteen job titles. There are four practical lanes for someone starting, and the titles sit inside them."}]},{"t":"p","runs":[{"text":"Analytics.","b":true},{"text":" SQL, spreadsheets, dashboards, and the ability to explain a number. Widest door, fastest to evidence, lowest ceiling early and a perfectly good ceiling later."}]},{"t":"p","runs":[{"text":"Data engineering.","b":true},{"text":" Pipelines, schemas, orchestration, reliability. More jobs than any other lane, more software engineering, longer runway to a first job unless you already program."}]},{"t":"p","runs":[{"text":"AI application engineering.","b":true},{"text":" Building products on top of models someone else trained: retrieval, APIs, evaluation, deployment. Newest lane, fewest settled conventions, most reachable if you can already write code and ship a web service."}]},{"t":"p","runs":[{"text":"AI-adjacent work that is not engineering.","b":true},{"text":" Product, operations, evaluation and annotation leadership, technical writing, solutions and customer-facing work, teaching. Real jobs, frequently overlooked, and often the best fit for someone with strong domain experience and no programming background."}]},{"t":"h2","id":"the-three-questions-that-pick-one","runs":[{"text":"The three questions that pick one"}]},{"t":"p","runs":[{"text":"Where do you already have context?","b":true},{"text":" Ten years in pharmacy, insurance claims, school administration or freight is an asset that a computer science graduate cannot acquire quickly. It points at a domain, and the domain often points at a lane: claims work points at analytics, freight at data engineering."}]},{"t":"p","runs":[{"text":"What can you evidence in eight weeks?","b":true},{"text":" Not learn: evidence. A finished, deployed, written-up thing."}]},{"t":"p","runs":[{"text":"What is actually hired where you are, or where you can legally work remotely?","b":true},{"text":" You measured this in the last lesson. If your city has forty analytics openings and two ML ones, that is not a conspiracy against your interests; it is the shape of your market."}]},{"t":"h2","id":"honest-time-to-first-evidence","runs":[{"text":"Honest time-to-first-evidence"}]},{"t":"p","runs":[{"text":"Part-time, around six hours a week, starting from ordinary computer literacy:"}]},{"t":"ul","items":[[{"text":"Analytics:","b":true},{"text":" eight to twelve weeks to a defensible project. SQL is learnable to working level in weeks."}],[{"text":"AI application engineering:","b":true},{"text":" three to four months if you already program, eight or more if you do not."}],[{"text":"Data engineering:","b":true},{"text":" four to six months, because it needs programming plus systems plus at least one orchestration tool."}],[{"text":"Research:","b":true},{"text":" years, and a different path entirely."}]]},{"t":"p","runs":[{"text":"These are ranges from watching many people do it, not guarantees. Some move faster. Almost nobody moves faster than half of these."}]},{"t":"h2","id":"the-adjacency-map","runs":[{"text":"The adjacency map"}]},{"t":"p","runs":[{"text":"The reason this choice is less frightening than it feels: the lanes touch."}]},{"t":"ul","items":[[{"text":"Support, operations or finance → analyst. Very common."}],[{"text":"Analyst → analytics engineer → data engineer. Very common."}],[{"text":"Data engineer → ML engineer. Common, and much easier than the reverse."}],[{"text":"Backend engineer → AI application engineer. Short hop."}],[{"text":"Analyst → product. Common at three to five years."}],[{"text":"Anything → research. Rare without a formal route."}]]},{"t":"p","runs":[{"text":"Notice what is missing: a direct jump from courses to ML engineering. It happens, but it is the exception people write posts about, which is exactly why it looks frequent."}]},{"t":"h2","id":"what-transfers-when-you-move-and-what-does-not","runs":[{"text":"What transfers when you move, and what does not"}]},{"t":"p","runs":[{"text":"Transfers: the domain you learned, the habit of finishing, your written record of work, SQL, debugging, and the thirty people who have seen you work."}]},{"t":"p","runs":[{"text":"Does not transfer: certificates, course completions, and tool-specific fluency in a tool the next team does not use."}]},{"t":"p","runs":[{"text":"The practical consequence is that the switching cost is a project and a job change (six to eighteen months) not a restart. People treat a lane choice as though it were a degree subject, then delay for months rather than pick. The delay costs more than a wrong pick would have."}]},{"t":"h2","id":"the-failure-mode-that-wastes-a-year","runs":[{"text":"The failure mode that wastes a year"}]},{"t":"p","runs":[{"text":"Studying three lanes at thirty per cent each. It feels prudent and it is the single most common way a promising year produces nothing hireable, because employers hire on depth in one thing and nobody is hired for breadth at zero years."}]},{"t":"p","runs":[{"text":"A workable rule: "},{"text":"one lane for six months.","b":true},{"text":" Then reassess on evidence, not mood: did you finish something, did anyone use it, did anyone reply to you about it. Boredom in week three is not evidence."}]},{"t":"h2","id":"what-to-do-if-you-cannot-get-any-job-at-all","runs":[{"text":"What to do if you cannot get any job at all"}]},{"t":"p","runs":[{"text":"Then the lane is chosen for you, and that is fine. Take the job that exists, then use the bridge-role method in the next module to move toward data from inside. Choosing a lane is a luxury of having options, and building options is the actual first task."}]},{"t":"h2","id":"the-limitation","runs":[{"text":"The limitation"}],"role":"limit"},{"t":"p","runs":[{"text":"None of this survives contact with an unusual opportunity. If someone offers you paid work in a lane you did not choose, take it. Paid work in the wrong lane beats unpaid preparation in the right one, every time, because it converts you from a candidate into someone with a reference."}]}],"check":{"question":[{"text":"Rahel has eleven years in hospital administration, no programming, six hours a week, and wants out of admin within a year. She is deciding between teaching herself deep learning and teaching herself SQL and reporting. Which reasoning is soundest?"}],"options":[{"text":[{"text":"Deep learning, because AI roles pay more and hospital data is exactly what modern models are built for."}],"correct":false,"tempting":[{"text":"Chooses by destination salary rather than by time-to-evidence, and ignores that the deep learning route from zero takes years at six hours a week."}]},{"text":[{"text":"Neither yet: she should first complete a broad foundation covering statistics, programming, machine learning and data engineering."}],"correct":false,"tempting":[{"text":"The breadth-first plan, which feels responsible and reliably produces a year with nothing finished and nothing hireable."}]},{"text":[{"text":"SQL, but she should leave hospital work behind entirely, since her admin experience will not count in a technical role."}],"correct":false,"tempting":[{"text":"Discards the domain knowledge that is the least replaceable thing she owns."}]},{"text":[{"text":"SQL and reporting, because it is the lane where she can produce a finished, defensible piece of work within about three months, and her eleven years of hospital context makes her a stronger candidate for health analytics than a generalist with better maths."}],"correct":true,"tempting":null}],"order":[0,1,3,2],"why":null},"prev":{"slug":"degrees-and-doors","title":"Who actually needs a degree","minutes":6},"next":{"slug":"the-ladder-and-the-levels","title":"Levels, titles, and why “senior” means nothing on its own","minutes":8}}}