Learning Library

← Back to Library

Video 45QmLivYv3k

Key Points

  • A McKinsey study reports that developers can finish coding tasks up to twice as fast when using generative AI, especially for repetitive, low‑complexity work.
  • Productivity is measured not by lines of code but by delivery-oriented metrics such as DORA (deployment frequency, lead time, MTTR) and project‑management tools like Jira.
  • While generative AI markedly boosts team productivity and developer experience, it shows little impact on complex coding tasks, indicating that human developers are still essential.
  • AI streamlines workflows through three main capabilities: automating repetitive coding tasks, offering natural‑language interfaces for code generation and debugging, and providing advanced code‑suggestion/autocomplete that helps overcome unfamiliar APIs and “coder’s block.”
  • The optimal outcome is a hybrid model where human developers collaborate with generative AI, leveraging these tools to accelerate routine work while retaining human expertise for complex problem‑solving.

Full Transcript

# Video 45QmLivYv3k **Source:** [https://www.youtube.com/watch?v=45QmLivYv3k](https://www.youtube.com/watch?v=45QmLivYv3k) **Duration:** 00:07:47 ## Summary - A McKinsey study reports that developers can finish coding tasks up to twice as fast when using generative AI, especially for repetitive, low‑complexity work. - Productivity is measured not by lines of code but by delivery-oriented metrics such as DORA (deployment frequency, lead time, MTTR) and project‑management tools like Jira. - While generative AI markedly boosts team productivity and developer experience, it shows little impact on complex coding tasks, indicating that human developers are still essential. - AI streamlines workflows through three main capabilities: automating repetitive coding tasks, offering natural‑language interfaces for code generation and debugging, and providing advanced code‑suggestion/autocomplete that helps overcome unfamiliar APIs and “coder’s block.” - The optimal outcome is a hybrid model where human developers collaborate with generative AI, leveraging these tools to accelerate routine work while retaining human expertise for complex problem‑solving. ## Sections - [00:00:00](https://www.youtube.com/watch?v=45QmLivYv3k&t=0s) **AI Doubles Coding Speed, Metrics Explained** - The speaker discusses a McKinsey study claiming generative AI can double developers' coding speed, explores how this might happen, how productivity is measured—including metrics like DORA—and argues AI will boost but not replace human developers. ## Full Transcript
0:00so listen to this a McKenzie study 0:02claims that software developers can 0:04complete coding tasks up to twice as 0:06fast with the help of generative AI 0:10which at least to me brings to mind 0:13three questions and number one 0:15is how how can it do that and we'll get 0:19to that in a moment because I have 10 0:21hows right here secondly I'd really like 0:25to understand how does anybody actually 0:28measure this stuff and anyway how do we 0:30know how productive somebody actually is 0:34and then lastly does this mean that 0:36generative AI could work even faster if 0:39we did away with those human software 0:43developers 0:45entirely well no human developers are 0:49not going anywhere anytime soon humans 0:53we still need you the researchers found 0:55that complex coding tasks were not 0:57significantly affected by the use of 0:59generative AI so worries about AI 1:02replacing developers can be safely laid 1:05to rest for now but generative AI did 1:08drastically speed up team productivity 1:10and improve the developer experience in 1:12a number of lwh hanging fruit use cases 1:14now as for actually measuring developer 1:17productivity so we can tell if 1:18generative AI has helped or hindered 1:20things well you could measure 1:22productivity based on volume how many 1:25lines of code were written but I think 1:28we can all agree that more does doesn't 1:30necessarily mean better I mean if 1:33somebody wants to pay me by the length 1:34of this video I could prattle on for 1:37hours but I don't think any of us would 1:38be feeling particularly productive after 1:40that so fortunately we have metrics 1:44metrics like Dora that stands for devops 1:49research and assessment and that has 1:51metrics like deployment frequency lead 1:53time and meantime to recovery for 1:56evaluating the efficiency of software 1:58delivery and we have project management 2:00tools like jira and those can track 2:04progress manage tasks and facilitate 2:06contribution analysis and what this is 2:09telling us is that productivity Peaks 2:12when we combine human software 2:14developers with generative AI so let's 2:18consider some ways that AI can 2:21streamline development workflows and 2:24number one up here is firstly to 2:27eliminate repetitive task Tas s coding 2:31often involves simple sometimes tedious 2:33tasks and this is where generative AI 2:36tools tend to shine repetitive routine 2:39work like typing out and customizing 2:41standard functions can be expedited 2:44where generative AI simply creates that 2:46code for us and for that to happen we 2:49can use method number two which is the 2:53natural language interface of large 2:56language models so we can ask for the 2:58generation of code snippet FS or debug 3:01code and provide Version Control simply 3:04by asking the AI model nicely in plain 3:06English or whatever our human language 3:09of choice is now number three that is 3:14code suggestion now this goes far beyond 3:17the code assist we've had for years 3:19that's essentially autocomplete but with 3:21code suggestion an AI can help you get 3:24started with an unfamiliar library or an 3:26unfamiliar package or if you're simply 3:29suffering from from coder's block 3:31staring at a blank screen in vs code for 3:35hours well generative AI can get you 3:38going and into the flow of coding sooner 3:41and once you do have some code created 3:43generative AI can then suggest number 3:46four code improvements and it does this 3:50by identifying redundant or inefficient 3:52sections if done right it's like having 3:54a peer programmer sat alongside you 3:57except without the skeptical looks and 4:00number five is code translation now this 4:03is where generative II can translate 4:05code from one language to another 4:07language which could be very useful in 4:09let's say an application modernization 4:11project where we might need to convert 4:12cobal Cod to let's say Java for instance 4:16now I have some more but before we go 4:19into what those are these 6 through 10 4:23let's really consider how generative AI 4:25is able to perform these tasks at all 4:28and it all starts essentially with a 4:31phase called pre trining now pre trining 4:38is what we do with our generative AI 4:41models and we take massive data sets 4:43containing diverse examples of code 4:45written in various programming languages 4:47and we feed it into this pre-training 4:49model the model learns to predict the 4:51the next word or the next token and a 4:53sequence of code based on the context of 4:55the preceding words and this allows the 4:57model to capture the syntax the 4:59semantics and the patterns inherent in 5:00different programming languages now when 5:03it comes to inference time here this is 5:06where we provide a prompt or a query 5:09into the generative AI model and it 5:10processes that input and uses its 5:12learned knowledge to understand the 5:16context and the intent of what we're 5:18asking for now the model considers all 5:20of the relationships between the 5:22different code elements such as the 5:24variables the functions and the control 5:25structures to generate relevant and 5:27syntactically correct code and then 5:30using the Learned patterns and 5:31contextual understanding the generative 5:32AA model can then generate for us an 5:35actual code 5:37snippet that contains the code that we 5:40asked for as the output and this code is 5:43based on the input prompt and follows 5:45the structure and style of the 5:47programming languages which the model 5:48was trained on and this isn't a closed 5:51loop because the model can then adapt to 5:54user feedback helping the model to 5:57refine its understanding and improve 5:59future 6:00outputs okay back to the way that 6:03generative AI can help with coding now 6:06where did we get to yeah number six so 6:08number six that is code testing now it 6:12can create test cases by analyzing code 6:15and generating test inputs and then even 6:18run those inputs and evaluate the 6:19resulting outputs in a similar vein AI 6:23can number seven perform bug detection 6:27in identifying and even automatic fixing 6:30bugs AI can also help create 6:33personalized Dev environments that adapt 6:37to individual developer preferences and 6:39coding Styles and in a personal favorite 6:42AI can be used to generate 6:45documentation so I don't have to by 6:48summarizing code functionalities 6:50explaining algorithms and providing 6:52context look well generative AI encoding 6:55is a powerful tool it is not a 6:57substitute for the creativity problem 7:00solving and domain expertise of human 7:02developers we should think of it as an 7:04augmentation tool something that assists 7:07developers in coding tasks providing 7:09suggestions and potentially speeding up 7:11certain aspects of the development 7:14process oh and no no no I haven't 7:16forgotten haven't forgotten number 7:1810 because that one's for you how has 7:22generative AI improved your coding 7:24experience is there something I haven't 7:26mentioned here let me know in the 7:28comments 7:30if you like this video and want to see 7:31more like it please like And 7:34subscribe if you have any questions or 7:36want to share your thoughts about this 7:38topic please leave a comment below to 7:40learn more please reach out to your IBM 7:42sales team or IBM business partner