{
  "$type": "site.standard.document",
  "content": "---\ntitle: \"So this is how it feels when the robots come for your job\"\ndescription: \"GitHub Copilot is a genuine force multiplier for coding, but human expertise is still crucial---a Conversation article on what AI assistants mean for programmers.\"\ntags:\n  - dev\n  - ai\n---\n\nimport Picture from \"@/components/Picture.astro\";\n\n<p class=\"post-subtitle\">what GitHub's Copilot 'AI assistant' means for coders</p>\n\n<Picture\n  file=\"posts/file-20220629-24-n3q489.webp\"\n  alt=\"a person typing away at some code on a laptop computer\"\n/>\n\n:::info\n\nThis article originally\n[appeared in the Conversation](https://theconversation.com/so-this-is-how-it-feels-when-the-robots-come-for-your-job-what-githubs-copilot-ai-assistant-means-for-coders-185957).\n\n:::\n\nI love writing code to make things: apps, websites, charts, even\n[music](/livecoding/). It's a skill I've worked hard at for more than 20 years.\n\nSo I must confess\n[last week's news](https://github.blog/2022-06-21-github-copilot-is-generally-available-to-all-developers/)\nabout the release of a new \"AI assistant\" coding helper called\n[GitHub Copilot](https://copilot.github.com) gave me complicated feelings.\n\nCopilot, which spits out code to order based on \"plain English\" descriptions, is\na remarkable tool. But is it about to put coders like me out of a job?\n\nTo understand what Copilot actually is, start with where it came from.\n[GitHub](https://github.com/features/copilot/#faq-human-oversight) (now\n[owned by Microsoft](https://news.microsoft.com/2018/06/04/microsoft-to-acquire-github-for-7-5-billion/))\nis a collaboration platform and social network for coders. You can think of it\nas something like a cross between Dropbox and Instagram, used by everyone from\nindividual hobbyists through to highly paid software engineers at big tech\ncompanies.\n\nOver the past decade or so, GitHub's users have uploaded tens of billions of\nlines of code for more than 200 million apps. That's a lot of `if`s and `for`s\nand `print(\"hello world\")` statements.\n\nThe Copilot AI works like many other machine learning tools: it was \"trained\" by\nscanning through and looking for patterns in those tens of billions of lines of\ncode written and uploaded by members of GitHub's coder community.\n\nThe training can take many months, hundreds of millions of dollars in computing\nequipment, and enough electricity to run a house for a decade. Once it's done,\nthough, human coders can then write a description (in plain English) of what\nthey want their code to do, and the Copilot AI helper will write the code for\nthem.\n\nBased on the [Codex \"language model\"](https://openai.com/blog/openai-codex/),\nCopilot is the next step in a long line of \"intelligent auto-completion\" tools.\nHowever, these have been far more limited in the past. Copilot is a significant\nimprovement.\n\nIn practice, it's a startlingly effective assistant. I was given early\n\"preview\" access to Copilot about a year ago, and I've been\nusing it on and off. It takes some practice to learn exactly how to frame your\nrequests in English so the Copilot AI gives the most useful code output, but it\ncan be startlingly effective.\n\nHowever, we're still a _long_ way from \"Hey Siri, make me a million dollar\niPhone app\". It's still necessary to use my software design skills to figure out\nwhat the different bits of code should do in my app.\n\nTo understand the level Copilot is working at, imagine writing an essay. You\ncan't just throw the essay question at it and expect it to produce a useful,\nwell-argued piece. But if you figure out the argument and maybe write the topic\nsentence for each paragraph, it will often do a pretty good job at filling in\nthe rest of each paragraph automatically.\n\nDepending on the type of coding I'm doing, this can sometimes be a huge time-\nand brainpower-saver.\n\nThe effectiveness does come with some open questions, though. I'm a\nbit worried they'll introduce, and reinforce, winner-takes-all dynamics: very\nfew companies have the data (in this case, the billions of lines of code) to\nbuild tools like this, so creating a competitor to Copilot will be challenging.\n\nAnd will Copilot itself be able to suggest new and better ways to write code and\nbuild software? We have seen AI systems\n[innovate](https://www.wired.com/2016/03/two-moves-alphago-lee-sedol-redefined-future/)\nbefore. On the other hand, Copilot may be limited to doing things the way we've\nalways done them, as AI systems\n[trained on past data](https://www.wired.com/story/ai-biased-how-scientists-trying-fix/)\nare prone to do.\n\nMy experiences with Copilot have also made me very aware my expertise is still\nneeded, to check the \"suggested\" code is actually what I'm looking for.\n\nSometimes it's trivial to see that Copilot has misunderstood my input. Those are\nthe easy cases, and the tool makes it easy to ask for a different suggestion.\n\nThe trickier cases are where the code looks right, but it may contain a subtle\nbug. The bug might be because this AI code generation stuff is _hard_, or it\nmight be because the billions of lines of human-written code that Copilot was\ntrained on contained bugs of their own.\n\nAnother concern is\n[potential issues](https://fossa.com/blog/analyzing-legal-implications-github-copilot/)\nabout licensing and ownership of the code Copilot was trained on. GitHub has\nsaid it is\n[trying to address these issues](https://github.com/features/copilot/#faq-human-oversight),\nbut we will have to wait and see how it turns out.\n\nAt times, using Copilot has made me feel a little wistful. The skill I often\nthink makes me at least a _little bit_ special (my ability to write code and\nmake things with computers) may be in the process of being \"automated away\",\nlike many other jobs have been at different times in human history.\n\nHowever, I'm not selling my laptop and running off to live a simple life in the\nbush just yet. The human coder is still a crucial part of the system, but as\ncurator rather than creator.\n\nOf course, you may be thinking \"that's what a coder _would_ say\" ... and you may\nbe right.\n\nAI tools like Copilot, OpenAI's\n[text generator GPT-3](https://openai.com/blog/gpt-3-apps/), and Google's\n[Imagen text-to-image engine](https://imagen.research.google), have seen huge\nimprovements in the past few years.\n\nMany in white-collar \"creative industries\" which deal in text and images are\nstarting to wrestle with their fears of being (at least partially) automated\naway. Copilot shows some of us in the tech industry are in the same boat.\n\nStill, I'm (cautiously) excited. Copilot is a force multiplier in the most\noptimistic tool-building tradition: it provides more leverage, to increase the\nuseful output for the same amount of input.\n\nThese new tools and the new leverage they provide are embedded in wider systems\nof people, technology and environmental actors, and I'm really fascinated to see\nhow these systems reconfigure themselves in response.\n\nIn the meantime, it might help save my brain juice for the hard parts of my\ncoding work, which can only be a good thing.\n",
  "createdAt": "2026-05-13T23:14:48.530Z",
  "description": "GitHub Copilot is a genuine force multiplier for coding, but human expertise is still crucial---a Conversation article on what AI assistants mean for programmers.",
  "path": "/blog/2022/06/30/so-this-is-how-it-feels-when-robots-come-for-your-job",
  "publishedAt": "2022-06-30T00:00:00.000Z",
  "site": "at://did:plc:tevykrhi4kibtsipzci76d76/site.standard.publication/self",
  "tags": [
    "dev",
    "ai"
  ],
  "textContent": "GitHub Copilot is a genuine force multiplier for coding, but human expertise is still crucial---a Conversation article on what AI assistants mean for programmers.",
  "title": "So this is how it feels when the robots come for your job"
}