Mastering Engineering Flow with Windsurf

00:00:00.000 | Hey everyone, I'm Ishan, I'm an engineer here at WindSurf, and today we'll be talking a little

00:00:21.520 | bit about how to make the most out of your development experience, how to make the most

00:00:25.360 | out of this agenda flow that we all now have exposure to with our IDEs.

00:00:32.560 | And also talk a little bit about how these flows make us better developers.

00:00:38.560 | For those who are not aware, WindSurf is in the dev tool space.

00:00:42.540 | Our flagship product is the WindSurf editor, which is, oh, as you can see, okay, sorry,

00:00:52.600 | I can't even, will it not share if I, okay, all right.

00:00:59.140 | Our flagship product is the WindSurf editor, which is powered by our agent, Cascade, which

00:01:05.840 | was actually the first agentic IDE in the space.

00:01:08.420 | After we launched a lot of tools, a lot of similar products followed suit in that they realized

00:01:13.300 | that this agent experience was the way to go with this IDE.

00:01:19.840 | So, for those who aren't aware or haven't tried it out, highly suggest you try it out.

00:01:25.920 | But before we actually get into how WindSurf does agents, how we do flows, let's take a step

00:01:30.680 | back.

00:01:31.680 | Let's talk about some of the most iconic duos to date and those that have really won,

00:01:38.220 | that have been successful.

00:01:41.880 | What are some of the characteristics of these duos and what has really led them to be so successful?

00:01:48.040 | It wasn't that each individual or each one of these were the best at their position or they

00:01:53.380 | were the best by far or considered the GOAT, although some may argue otherwise for some of

00:01:58.320 | these people.

00:01:59.920 | What made them so good was that together as a team, they worked better than everybody

00:02:04.920 | else and that's what led to success.

00:02:09.140 | One of the most iconic duos, pineapple and pizza, regardless of what you all say.

00:02:14.920 | This is what makes duos successful is that they work together and they know each other.

00:02:19.760 | They complement each other very well.

00:02:22.680 | Another one of the most iconic duos.

00:02:24.600 | What this reminds me of is how we think of our agents or how we think of our coding assistants.

00:02:29.880 | A lot of times we fight with them.

00:02:34.460 | We ask ourselves, "Why is it not getting us?

00:02:36.500 | Why is it producing all this error-prone code?

00:02:39.440 | Why is it duplicating code?

00:02:40.520 | Why is it removing code that we don't want it to remove?"

00:02:43.880 | We also tend to abuse our agents sometimes.

00:02:47.420 | We tell it that we're going to fine it one million dollars if it doesn't do what we ask

00:02:50.440 | it to do.

00:02:51.440 | We also tell it that it's going to go to jail if it doesn't do what we want it to do.

00:02:55.720 | But that's not how it's supposed to be.

00:02:58.080 | We are supposed to treat our agents like we would treat our running mate, our teammate,

00:03:03.660 | our friend.

00:03:04.700 | And this is what it's supposed to look like.

00:03:06.280 | Right?

00:03:07.280 | Not like what Tom and Jerry are.

00:03:09.280 | So, it's not your fault as developers.

00:03:13.280 | It's ours.

00:03:14.280 | It's ours as the builders of these tools to provide that experience for you.

00:03:18.160 | To make sure that you as developers, you're actually able to work with these agents.

00:03:23.280 | that these agents understand you.

00:03:25.860 | How do we fix this problem?

00:03:26.860 | How does Windsurf approach this?

00:03:29.860 | Before I dive in depth into that, let's take a step back and look at how we've gotten to

00:03:36.180 | this point of agents, the evolution of AI, how AI has kind of evolved into what we see today

00:03:42.860 | with these coding agents.

00:03:44.900 | Back in 2022, the ancient times, we had to do everything ourselves, right?

00:03:51.480 | It's kind of terrifying.

00:03:52.480 | We had to write all the code by ourselves.

00:03:54.480 | There was no AI to really help us, right?

00:03:55.480 | There was just Stack Overflow and Google.

00:03:58.480 | So, in 2022 or late 2022 and moving on into '23, co-pilots were introduced.

00:04:05.980 | I consider ChatGPT barred at the time, GitHub co-pilot as this kind of chat interface, this

00:04:12.840 | co-pilot kind of experience where you would give a prompt or an input, you'd get a response.

00:04:19.540 | Really good with simple Q&A and autocomplete.

00:04:22.800 | Late 2024, this is when we saw the first agents, or what people like to call agents.

00:04:31.620 | A lot of times, people would confuse agents with workflows, but I think that was kind of

00:04:35.660 | cleared up more recently in that agents now are able to take this autonomy, they take this

00:04:41.040 | independence, they operate iteratively, they adjust their trajectories, and they can perform

00:04:47.160 | these larger scope tasks and do things that simple or single response, single shot co-pilots

00:04:54.380 | would not be able to do.

00:04:57.340 | So another way to look at this is that co-pilots were a little bit more collaborative, right?

00:05:00.560 | We had to interact at every single step of the process while we're working with co-pilots.

00:05:04.620 | If we ever were coding with ChatGPT, we'd have to maybe paste in every file, we would have

00:05:09.500 | to individually send in each response, and ChatGPT would have to work on things very step-by-step,

00:05:16.720 | right?

00:05:17.720 | So we would just send an input or a prompt, maybe add some extra context, and after the

00:05:22.160 | LLM ran its inference, we would get the response.

00:05:25.460 | Agents then introduced this autonomous nature where we had these models, we also allowed it

00:05:31.980 | to retrieve this context.

00:05:33.640 | But on top of that, we introduced tool calling and function calling, the ability for agents

00:05:37.400 | to perform tasks, perform these functions, and actually be able to execute things that

00:05:43.500 | co-pilots or chatbots wouldn't necessarily be able to do.

00:05:47.660 | What Windsurf has done when we launched was we realized that agents really didn't solve the

00:05:53.120 | problems that co-pilots couldn't solve, and co-pilots didn't do everything that we would

00:05:58.020 | want to do as developers as well.

00:05:59.920 | And so we took the best of both worlds, in that we combined these models, as well as

00:06:07.680 | retrieving proper context, figuring out how to call tools properly, in addition with understanding

00:06:14.540 | the user, tracking the user's actions, really understanding their intent as developers, what

00:06:20.320 | they're going to do.

00:06:22.340 | So we essentially took the best of both worlds, right?

00:06:26.120 | And so AI Flows is what Windsurf and Cascade introduced back in November.

00:06:30.840 | And this is where a lot of other companies, a lot of other products that I'm sure you all

00:06:34.440 | know today, followed suit in that they realized this simple chat interface, or even having an

00:06:38.960 | agent and a chat just maybe didn't make sense.

00:06:41.800 | It only made sense to provide one singular agentic interface that collaborated with you as a developer.

00:06:48.060 | And so Windsurf, our editor, it took the best of both worlds, in that it combined this collaborative

00:06:53.980 | power of a chatbot interface with some of these autonomous and tool calling capabilities of an

00:07:00.820 | agent, and molded them together to work in perfect unison, in perfect sync.

00:07:06.280 | And this created a very seamless, a very unified experience where developers and AI could actually

00:07:11.800 | operate as one, could operate as a team, rather than like these guys.

00:07:17.080 | So how did Windsurf do this?

00:07:21.120 | The first thing we wanted to consider was this concept of flow awareness.

00:07:25.460 | What this meant was that we would have really comprehensive reasoning and understanding of the

00:07:30.840 | implicit user intent.

00:07:32.280 | This is something that a lot of agents actually don't really consider right now, and even when

00:07:37.860 | we're building this on the side, we don't actually think of what is really important to the user.

00:07:42.160 | So our main emphasis when we were building Cascade was understanding the user, tracking their actions,

00:07:47.000 | their edits, their commands, the terminal commands they've run, anything that's in their clipboard,

00:07:54.040 | all the files that they've recently edited.

00:07:55.840 | All these things are tracked by Cascade, so that Cascade develops an understanding of what

00:08:00.840 | the user has been doing, and that is all inputted into this agent's trajectory.

00:08:05.820 | So that the agent can then outline a much more relevant set of steps that aligns with what

00:08:11.620 | the user would be doing, or what may be doing in the future.

00:08:15.300 | And so when you use Cascade, when you use Windsurf Tab, it almost feels like Windsurf is predicting

00:08:20.020 | your next step.

00:08:21.420 | It's almost like an LLM.

00:08:22.420 | It predicts the next token.

00:08:23.880 | In this case, Cascade is predicting or inferring what you would do next.

00:08:27.400 | When you're using Tab, it feels like Cascade or Windsurf is reading your mind because so

00:08:33.280 | much of this context of the user's actions as well as this understanding of the user over

00:08:39.580 | time as the user interacts with Cascade, this is inputted into the context window.

00:08:44.380 | And this is what makes this agent so powerful is that it understands the user much better than

00:08:49.780 | a very general purpose or independent agent.

00:08:53.480 | On top of that, we have a state-of-the-art context engine.

00:08:55.800 | We have a talk on this tomorrow as well.

00:08:57.900 | But we approach context differently than other products in this space.

00:09:00.960 | We don't just use a RAG or embedding-based search approach.

00:09:04.520 | We leverage a combination of multiple tools in tandem to really figure out what works best

00:09:10.180 | so that we can understand your code base, this explicit context, as strong and comprehensively

00:09:15.400 | as possible.

00:09:16.480 | And what this helps us do as an agent is it helps us get more accurate results, more relevant

00:09:21.880 | suggestions for your code base or for any of these code suggestions.

00:09:27.180 | We have reduced hallucinations.

00:09:29.000 | We don't have to do any guesswork.

00:09:31.380 | Because we understand the user, because we understand your code base very well, we are able to provide

00:09:36.000 | a much more tailored experience, an experience that actually outlines what the user may want.

00:09:41.340 | And it's grounded in this centralized source of truth, which is this code base that you're

00:09:46.280 | working in.

00:09:48.280 | On top of that, we equip Cascade with all of the best tools for this agent to really perform

00:09:53.840 | that multi-step iterative kind of set of tasks where it's not just limited to that single

00:09:59.740 | response nature.

00:10:00.840 | It can actually call MCP servers.

00:10:02.840 | It can leverage workflows, something we recently introduced, where now we are bringing together,

00:10:07.800 | bridging that gap between the unpredictability of an agent and more of that deterministic nature

00:10:13.260 | of a workflow.

00:10:13.840 | Workflow, right?

00:10:14.840 | Where workflows, we know exactly what's going to happen and when it's going to happen.

00:10:18.340 | And so with Windsurf, again, our main priority is to help developers as much as possible.

00:10:23.560 | We realize that with agents, it's very unpredictable what's going to happen and what they're going

00:10:27.220 | to do.

00:10:28.220 | And so with workflows now, this bridges that gap.

00:10:30.640 | We can actually, as users, define a set of steps for the agent to follow as they're operating.

00:10:36.840 | On top of that, we allow Cascade to look at rules, right?

00:10:40.580 | These are rules that you can generate as users.

00:10:43.360 | They could be file-based rules.

00:10:44.660 | They can be rules that you always want the model to look at.

00:10:48.560 | You maybe sometimes want the model to look at.

00:10:51.400 | On top of that, Cascade can generate memories of you as a user, some of your preferences, memories

00:10:56.840 | of your code base so that it doesn't have to constantly index, doesn't have to constantly

00:11:00.400 | retrieve your code base.

00:11:01.840 | And on top of that, recently, there's a lot of talk about ATA, A2AA, and multiple agents.

00:11:07.080 | Recently we introduced multiple simultaneous Cascades working in tandem.

00:11:11.040 | And so what this brings to us is it allows Cascade now to have these multiple different trajectories

00:11:18.740 | that understand each other, understand the user, and understand your code base, which allows

00:11:23.920 | us to be able to complete tasks faster, more efficiently, and get things done in the way

00:11:29.720 | that we want to get them done.

00:11:31.800 | So all these tools, right, there's a lot that I left out here, like the ability to search

00:11:36.520 | the web and other things like that.

00:11:38.340 | They empower Cascade to be able to really give us this agentic feeling while it's operating,

00:11:43.540 | right?

00:11:44.540 | And really, not just, again, an independent agent, but one that understands us as users.

00:11:49.320 | But as engineers, how do we make the most?

00:11:50.900 | How do we get the most out of these flows?

00:11:54.660 | Also known as engineer maxing, how do we really max out what we get out of this experience?

00:12:00.660 | The way I like to approach it, I'm sure you all are somewhat familiar with this kind of

00:12:03.740 | set of steps.

00:12:04.740 | We like to first explore, discover our code base, scope out our tasks, then plan, then build,

00:12:10.540 | code, and then test out our changes or anything that the agent did.

00:12:15.460 | When discovering, we can leverage Cascade at every set of these steps, right?

00:12:20.260 | We want to leverage Cascade to understand our code base, to scope out our tasks, determine

00:12:24.420 | this definition of done, right?

00:12:25.880 | That's what an agent does, right?

00:12:26.880 | It determines the definition of done and outlines a set of steps.

00:12:30.180 | How can we tell Cascade what that definition of done is and talk through some of these goals?

00:12:35.020 | On top of that, if we wanted to give more direct context, we cannot mention different

00:12:38.920 | files and directories.

00:12:41.160 | It's also very important to plan with Cascade, right?

00:12:43.500 | A lot of people just say, hey, I'm just going to put in this prompt and let Cascade or let

00:12:47.120 | this agent just build everything for me.

00:12:49.140 | The best way to go about this is create, work with the agent to understand you.

00:12:53.380 | And so what that means is create a planning file, outline a set of tasks with maybe check

00:12:58.800 | boxes to tell Cascade, hey, you need to hit these set of tasks to accomplish this definition

00:13:04.020 | of done.

00:13:05.020 | Also outline rules to tell Cascade how to behave that align with your preferences.

00:13:10.820 | And then we go into building.

00:13:12.380 | Work with Cascade to actually execute this plan that you've set for yourself.

00:13:16.580 | Edit files one by one or multiple files at a time and ensure Cascade is working with this

00:13:21.400 | planning document and actually checking all the boxes.

00:13:25.860 | And you'll see that Cascade actually asks you questions.

00:13:28.380 | It will check in with you as a user, as a developer, right?

00:13:31.260 | And you make sure to answer it very clearly.

00:13:33.660 | Tell it, hey, you're doing this incorrectly.

00:13:35.460 | Or, hey, let's actually look at this first.

00:13:37.460 | Or let's look at this set of steps next, right?

00:13:41.020 | WinServTab then provides this more hands-on experience with autocomplete.

00:13:44.820 | And on top of that, you can leverage these simultaneous Cascades and these different MCP servers that

00:13:50.020 | allow you to pull additional context and empower Cascade with these additional tools to complete

00:13:55.380 | this agentic experience.

00:13:57.820 | Next, lastly, then you want to generate tests with Cascade, right?

00:14:01.620 | And generate, run, fix your tests one by one as it's iterating.

00:14:06.180 | You can leverage workflows here, which you can leverage to automate some of these tasks.

00:14:12.620 | And at the end, determine if Cascade hit that DOD, that definition of done.

00:14:16.620 | If it didn't, make sure Cascade understands what it did wrong, because Cascade is building this learning of you.

00:14:22.420 | It's building this embedding representation of you as a developer, and it's learning from that.

00:14:27.220 | So in the future, it knows where to not go wrong and where to improve.

00:14:31.420 | And then lastly, then you want to commit to git, and make sure that you're pushing code that is functional.

00:14:38.820 | So you and Cascade, right, don't think of you guys as separate entities.

00:14:43.620 | We want to think of you all as a merge, as a team, right?

00:14:46.220 | As a peer programmer, you guys are working together to accomplish tasks.

00:14:50.420 | And so this is how you and Cascade should be.

00:14:52.820 | This is how developers and Cascade should be.

00:14:54.620 | It shouldn't be separate where you tell Cascade, "Hey, just accomplish this set of tasks.

00:14:58.420 | I'll come back in 30 minutes, and we'll see what you did," right?

00:15:02.920 | We're not at that level where LLMs can do that just yet.

00:15:05.520 | And so really work with Cascade.

00:15:07.620 | Make it your friend.

00:15:08.920 | And that is how you'll get the best development experience out of these agents.

00:15:13.020 | Thank you.

00:15:14.020 | We'll see you next time.

00:15:15.020 | We'll see you next time.

00:15:16.020 | We'll see you next time.

Mastering Engineering Flow with Windsurf - Eashan Sinha, Windsurf

Chapters