# Telling your agent “no” is a moat now, rearward deployed engineers, and harnessing the context for your SDLC | Dev Interrupted Powered by LinearB

> This week on the Friday Deploy, Ben and Andrew break down Uber's strategy of rearward deploying engineers to scale agentic workflows across non-technical teams. They also explore Claude Code making Auto Mode default, Meta's on-device Muse Glimmer model, Tim O'Reilly's open-source AI perspective, and the fundamentals of context engineering for your SDLC.

_This is a markdown rendering of a live HTML page on linearb.io, generated for AI/LLM consumption — it is not a markdown-only site. To get the full HTML page instead, request this URL with an explicit `Accept: text/html` header (no wildcard, no markdown preference)._


```json
{
  "@context": "https://schema.org",
  "@type": "PodcastEpisode",
  "name": "Telling your agent “no” is a moat now, rearward deployed engineers, and harnessing the context for your SDLC",
  "description": "This week on the Friday Deploy, Ben and Andrew break down Uber's strategy of rearward deploying engineers to scale agentic workflows across non-technical teams. They also explore Claude Code making Auto Mode default, Meta's on-device Muse Glimmer model, Tim O'Reilly's open-source AI perspective, and the fundamentals of context engineering for your SDLC.",
  "url": "https://linearb.io/dev-interrupted/podcast/rearward-deployed-engineers-claude-code-auto-mode-sdlc-context",
  "datePublished": "2026-08-14T12:00:00.000Z",
  "partOfSeries": {
    "@type": "PodcastSeries",
    "name": "Dev Interrupted",
    "url": "https://linearb.io/dev-interrupted/podcasts"
  },
  "actor": {
    "@type": "Person",
    "name": "Andrew Zigler",
    "jobTitle": "GTM Engineer",
    "worksFor": {
      "@type": "Organization",
      "name": "LinearB"
    }
  }
}
```

```json
{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Home",
      "item": "https://linearb.io/"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "Dev Interrupted - Podcasts",
      "item": "https://linearb.io/dev-interrupted/podcasts"
    },
    {
      "@type": "ListItem",
      "position": 3,
      "name": "Telling your agent “no” is a moat now, rearward deployed engineers, and harnessing the context for your SDLC",
      "item": "https://linearb.io/dev-interrupted/podcast/rearward-deployed-engineers-claude-code-auto-mode-sdlc-context"
    }
  ]
}
```

[Home](https://linearb.io/)

/

[Podcast](https://linearb.io/dev-interrupted/podcasts)

/

Telling your agent “no” is a moat now, rearward deployed engineers, and harnessing the context for your SDLC

# Telling your agent “no” is a moat now, rearward deployed engineers, and harnessing the context for your SDLC

By Andrew Zigler

|

August 14, 2026

![ai_agent_workflows_engineering_context_management_82ca8af964](https://assets.linearb.io/image/upload/c_limit,w_3840/f_auto/q_auto/v1/ai_agent_workflows_engineering_context_management_82ca8af964?_a=BAVMn6ID0)

This week on the Friday Deploy, Ben and Andrew explore Uber's strategy of "rearward deploying" engineers to spread agentic AI workflows into departments like legal and marketing. They also dive into Anthropic making Claude Code's Auto Mode the default, Meta's new on-device Muse Glimmer model, and Tim O'Reilly's case for an open source AI ecosystem. Finally, they break down context engineering for the SDLC and examine new research showing why generalized agent skills outperform personalized ones.

### Show Notes

* [After starting the tokenmaxxing panic, Uber's CTO is back with a very different AI story](https://www.businessinsider.com/uber-turns-best-ai-engineers-loose-pods-business-2026-8)
* [Auto mode is now the default in Claude Code for Pro, Max, and Team plans](https://claude.com/blog/auto-mode-default-in-claude-code)
* [Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device](https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model)
* [Why Open Source Matters for AI](https://oreillyradar.substack.com/p/why-open-source-matters-for-ai)
* [Your SDLC is your context engineering](https://leaddev.com/software-quality/your-sdlc-is-your-context-engineering)
* [Do personalized skills help coding agents? An empirical study of developer interaction histories](https://arxiv.org/abs/2608.10319)

### Transcript 

_(Disclaimer: may contain unintentionally confusing, inaccurate and/or amusing transcription errors)_

\[00:00:00\] **Ben Lloyd Pearson:** Ah, so Andrew, we got a new token, life beyond token maxing story yet again for this week. I think that's two weeks in a row now. Uh, but this is a different kind of life beyond token maxing story than what we're typically used to. You know, we're used to like people like revolting against token maxing and then putting all these restrictions on it.

\[00:00:19\] **Ben Lloyd Pearson:** But here we have Uber is back in the news, who's, you know, fame- known as one of the organizations that token maxed earlier this year. Uh, their CTO's out there sharing some new ideas on how they're deploying their agents, their, their engineers tokens now. And it's actually a really interesting idea. Like they're, they've, they're now sending their engineers out into other teams within the company, within Uber, out to legal, into marketing, into sales, and, um, they're calling them, uh, a rearward or rear deployed engineers rather than like a forward deployed engineer, which is kind of, uh, interesting.

\[00:00:54\] **Ben Lloyd Pearson:** Uh, we used to call it just like inner source, you know. Um, but yeah, it's a pretty interesting \[00:01:00\] idea of like using your engineers to, you know, you got all these productivity gains coming into your engineering organization. Why not have them just help, uh, people in other sides of the company capture those gains?

\[00:01:10\] **Ben Lloyd Pearson:** I, I don't know. What'd you think when you read this story, Andrew?

\[00:01:13\] **Andrew Zigler:** Well, I, I just thought that we really needed to workshop rearward deployed engineer, you know, maybe before going forward with that one, or

\[00:01:21\] **Ben Lloyd Pearson:** Yeah.

\[00:01:21\] **Andrew Zigler:** with that one in this case. So,

\[00:01:23\] **Ben Lloyd Pearson:** Yeah

\[00:01:23\] **Andrew Zigler:** was an interesting turn of events. You're right that, you know, we've had been on a little bit of a, a streak lately with week after week there's been some major company we've talked about or covered here, uh, indulging in the token maxing phenomenon, and then dramatically shifting course, either reversing it and going for, like, a very minimizing or cost restrictive approach, or in this case, realizing the more substantial opportunity.

\[00:01:47\] **Andrew Zigler:** And that is that if you have engineers that are able to access and work with that amount of tokens and deliver that much amount of value, maybe you're still figuring out what that value is, but if it's clearly there in some form, let's distribute \[00:02:00\] it. Let's figure out how to bring these gains into other departments that aren't as enabled, that don't have these, like, technical

\[00:02:07\] **Ben Lloyd Pearson:** Yeah

\[00:02:07\] **Andrew Zigler:** this is the same idea of, like, hiring an agency to come in and do, like, your AI transformation, except in this case, you're enlisting your smartest and most, like, natively and, uh, familiar, uh, folks for your ecosystem. So it's a really smart play. Um, I think this is how you actually kind of get the organizational gains, and that's gonna be a theme, I think, actually, across all the stuff we talk about in today's episode is how do you distribute this gains to work on an organizational level, and what does that mean?

\[00:02:35\] **Andrew Zigler:** I think that's the big challenge, so it's exciting to see a leader like Uber really take to the charge on it.

\[00:02:40\] **Ben Lloyd Pearson:** Yeah, and, and, and speaking of this theme, I mean, you know, we've, we've heard similar stories from LinearB customers where they've, you know, they've been this agentic leader within their organization. They've enabled the entire engineering organization to leverage agents and move faster than ever before.

\[00:02:55\] **Ben Lloyd Pearson:** And then the first question they get is like, "Hey, can you go help other teams learn how to do that \[00:03:00\] too, outside of engineering?" Uh, so, you know, I kinda think that this probably won't be common practice for too long. Um, I think it's mostly gonna be like larger enterprises where you see this, this sort of behavior pop up or companies that are just really, really far ahead on the agentic curve.

\[00:03:15\] **Ben Lloyd Pearson:** You know, 'cause at the end of the day, eventually people are gonna build tools for those other teams to solve their work agentically. Um, but you know, like I said, this is a different kind of life after token maxing that, that I'm, I'm totally here for. Uh, and just shameless plug, if you haven't listened to it yet, you know, we had, Angie and I hosted a workshop with LinearB a couple months back where we talked about this concept of token maxing and what it means to get past that and what, what life looks like once you've sort of moved beyond just looking at raw adoption and AI usage and start to think about where's that impact actually hitting your organization.

\[00:03:49\] **Ben Lloyd Pearson:** So yeah, cool little story. But anyways, welcome to the Friday Deploy, brought to you by LinearB. I'm your host, Ben Lloyd Pearson

\[00:03:59\] **Andrew Zigler:** And I'm your \[00:04:00\] host, Andrew Zigler

\[00:04:01\] **Ben Lloyd Pearson:** And this week, in addition to token maxing, we are covering model updates from Claude and Meta, why open source still matters, context engineering in your SDLC, and some research that answers the question, do personalized skills help coding agents? So let's just start right at the top with this news from Anthropic.

\[00:04:21\] **Ben Lloyd Pearson:** What do we-- what, what's all this about Claude Code, uh, now being defaulted to auto mode?

\[00:04:26\] **Andrew Zigler:** All right, so Claude Code, they're talking about auto mode here. Auto mode is something that, uh, Anthropic has very famously been tinkering with since really Claude Code hit the scene. The idea that the model could, uh, check its own permissions on commands that it runs on your own machine is a really important guardrail and restriction for working with the tool, and there's been all different layers in this discussion about where should that guardrail live, who should be in charge of it?

\[00:04:48\] **Andrew Zigler:** But Anthropic has really taken the mantle here in making these, uh, very safe evaluations a premier front part, uh, experience of using Claude Code. So auto mode has, uh, \[00:05:00\] been upgraded to being a default permission setting. Up until now, it was a experimental setting that you had to turn on, and the idea is that it catches pretty much any kind of red flag command that would typically, uh, you know, be a potentially destructive or harmful one.

\[00:05:16\] **Andrew Zigler:** I think these are kinds of rewards they've earned out of all of the work they've had to do in controlling and maintaining like the Mythos and Fable rollouts

\[00:05:26\] **Ben Lloyd Pearson:** Yeah, absolutely

\[00:05:27\] **Andrew Zigler:** cybersecurity abilities. They just now have so much intelligence around how to construct and create these boundaries that it makes sense they're gonna make this a default. I mean, folks like anybody who's, uh, already approving their prompts manually in Claude Code, they approve 97% of them, uh, just automatically anyways. This is actually a,

\[00:05:49\] **Ben Lloyd Pearson:** Linear

\[00:05:49\] **Andrew Zigler:** guardrail in many of those cases because it's paying a little bit of a closer attention. They also worked with some third-party red teams to try and do prompt injection on it, did a whole bunch of testing, \[00:06:00\] uh, and was able to stamp out, uh, you know, uh, harmful commands that had popped up in previous generations.

\[00:06:06\] **Andrew Zigler:** So really promising frontier research coming from Anthropic about how to protect a model from harming your system. Definitely a really, uh, and critical part of the model hosting infrastructure, especially as you move that stuff onto your own systems.

\[00:06:21\] **Ben Lloyd Pearson:** Yeah. You know, I think overall this change is probably a net benefit. Like, you know, I, I expect there's probably a lot of like pessimists out there that might be looking at the, the potential downsides to this where, you know, Anthropic could use this as a way to just route all of your prompts to cheaper models to save them on costs.

\[00:06:38\] **Ben Lloyd Pearson:** Uh, which is probably true. That's probably gonna happen to some degree But I think there's a real upside here too, in that, you know, Anthropic can also choose when to do things like implement longer thinking horizons or to use reasoning or to maybe use a better model instead of a cheaper model. Uh, you know, and when I'm using Claude, um, you know, I don't always wanna have to be forced to think \[00:07:00\] about whether or not I need all of these different capabilities and when it actually executes those capabilities, what...

\[00:07:05\] **Ben Lloyd Pearson:** Is it gonna be doing it securely? Um, so yeah, there was a couple of quotes that I, that I wanted to call out specifically from this article. Um, the first is about how it, it, you know, this tool or this, this ability routes each tool call through a classifier that targets blocking actions that are irreversible, destructive, or aimed outside your environments.

\[00:07:26\] **Ben Lloyd Pearson:** They have these new safeguards in place that, uh, you know, one of the biggest things that I'm always paranoid about when I'm, when I try to start giving Claude a little more freedom to take action on my tools is, um, y- you know, am I comfortable with the, uh, b- the permissions and the, the changes that it's making to the things that I'm connecting to it?

\[00:07:46\] **Ben Lloyd Pearson:** Um, and as you mentioned, the other thing that stood out, you know, th- they, they have data that suggests that, um, manual reviews are just habitual for most users. Uh, you mentioned the number 97%, so that means 97% of these tool call permissions that \[00:08:00\] Claude Code asks for just get approved, which seems to indicate that, you know, people were just clicking through reflexively rather than reviewing every command.

\[00:08:08\] **Ben Lloyd Pearson:** And I try to be conscious about when I-- what, what permissions I'm giving it, but, you know, I'm not gonna lie, there are times where I see it saying it's gonna make an API request, and I only skim what it's doing and don't s- actually, like, really investigate and look closely at what it's doing. But,

\[00:08:26\] **Andrew Zigler:** Totally

\[00:08:26\] **Ben Lloyd Pearson:** on this show for a while now that, you know, these foundational AI tools need better protections and safeguards like this that help protect us from our, our AI going rogue.

\[00:08:36\] **Ben Lloyd Pearson:** So I think it's really great to see that, you know, Anthropic is continuing to think about this challenge and is building tooling. Um, you know, I will say probably my biggest concern, and where, you know, I do align a little bit with some of the pessimists out there, is that we really do need the ability to turn things like this off when we need to, and be able to manually configure as much as possible.

\[00:08:57\] **Ben Lloyd Pearson:** You know, and I, and I keep thinking about this Hugging \[00:09:00\] Face OpenAI hack that just happened, and it really kind of is the perfect example of how AI safeguards can actually backfire on you. Uh, you know, in this example, Hugging Face was unable to use some of the Mythos class models for responding to the security breach in real time, um, because those models couldn't determine if, if the people pr- trying to protect themselves were being malicious or not.

\[00:09:24\] **Ben Lloyd Pearson:** Like, it actually looked like they could be potential bad actors themselves. Uh, so it would refuse to do things for them and, you know, and then the team had to switch to other models that just don't have those safeguards so that they could respond to this security incident. But yeah, there's a lot of benefits that Anthropic is claiming here.

\[00:09:40\] **Ben Lloyd Pearson:** You know, it supposedly reduces the risk of prompt injection, harmful action. So yeah, I think this is just sort of like the next step in what is likely to be a constant iteration of better safeguards around AI.

\[00:09:53\] **Andrew Zigler:** Yeah

\[00:09:55\] **Ben Lloyd Pearson:** All right. Let's talk about the latest news from Meta. This is a pretty cool announcement, I \[00:10:00\] feel like.

\[00:10:00\] **Ben Lloyd Pearson:** Um, Muse Glimmer. It's an open agentic model that runs on your devices. What, what is this, Andrew?

\[00:10:07\] **Andrew Zigler:** Yeah, this is, uh, like you described, it's a long-running agent coming from the Meta lab. So this is an open source, open weight model. If you're familiar with the, the, the models that they release, they do so fully, fully open source because they want a fully collaborative ecosystem. And this latest one is a 30 billion parameter open weight model an Apache 2 license, which we've talked about since around April this year.

\[00:10:29\] **Andrew Zigler:** That's becoming the really popular trend ever since around Gemma 4 of, of this type of a license, which means that you can fine-tune, train this, uh, make your own custom private model and sell services off of it. Uh, and there's no cloud dependency required either. The idea is that it can run on local hardware.

\[00:10:45\] **Andrew Zigler:** Um, it can even be on consumer grade GPUs. And I myself, I haven't had a chance to tinker with it yet, but I'm definitely very curious to give it a try on some of the machines I have around, and, um, I do think that owning your inference and having this long-running agent \[00:11:00\] is a really, uh, powerful and useful tool for folks, um, especially because this one is more focused on doing tasks and, uh, is, is not like a coding agent, right?

\[00:11:10\] **Andrew Zigler:** So this is a really great candidate for if you have hardware and you want to run a long-running, um, agentic assistant, um, especially one that lives on device or works with sensitive or private data. This is, becomes a really great candidate, um, for that kind of world. Of course, if you're using a model like this, you have to bring your own everything, including like your harness and the environment it's going to work in, and this comes back to what we just talked about a moment ago of things like guardrails.

\[00:11:36\] **Andrew Zigler:** You know, if you're going to use and you're going to ultimately have it working or, or, or, uh, operating on tasks, you have to think about the guardrails you have to bring to the system to make sure it operates safely, uh, within your environment. Of course, this is all just baseline stuff for working with any model, uh, but particularly important when you get long-running ones that live on your own device.

\[00:11:56\] **Andrew Zigler:** Really cool development. I'm excited to see what people build with it. \[00:12:00\] Um, what, what do you, what do you think of the latest developments?

\[00:12:03\] **Ben Lloyd Pearson:** Yeah. Well, first of all, I'll, I'll love to hear what you think after you get it into your lab and dissect it and benchmark it and see what your agents think about it. Um, but, you know, also I would love to hear, you know, friend of show, Brigida Bowkler, I would love to hear her opinion on this too, 'cause I know we just recently covered some research she's been doing around the viability of local models, and at the time her conclusion was that they still needed some time to develop, like they weren't quite it yet.

\[00:12:28\] **Ben Lloyd Pearson:** Um, but there was definitely potential that seems like it's on the immediate horizon. So, uh, that's pretty cool. I'd really, I'm really looking forward to... I, I know she's probably out there already thinking about this. Um,

\[00:12:39\] **Andrew Zigler:** A cool note for this one too is that this one was mostly trained via distillation

\[00:12:43\] **Ben Lloyd Pearson:** I know, yeah

\[00:12:44\] **Andrew Zigler:** model, teacher model. That's a really important note I wanna call out for our listeners because

\[00:12:48\] **Ben Lloyd Pearson:** Mm-hmm.

\[00:12:48\] **Andrew Zigler:** the, that's the trend for all of these open source models, is that, uh, you get these loops where they're trained or, or, uh, created from synthetic data that's constructed by a smarter model or by like a \[00:13:00\] more, uh, frontier model.

\[00:13:01\] **Andrew Zigler:** So right now, a lot of the benefits we get in the open source world are just coming off of like the comet trails, right, of these foundation models in a sense.

\[00:13:10\] **Ben Lloyd Pearson:** Yeah. Yeah. And, and my impression that th-this, this, this model specifically is sort of part of Meta's goal to attach AI to your desktop work environment. So, you know, we've probably been, we've been hearing all these stories about how Meta's tracking their employees', uh, computer usage and using it to train, uh, some new models.

\[00:13:28\] **Ben Lloyd Pearson:** Uh, so I, I imagine that that's really what this is, what has been used to sort of get this model to where it is today or one of the ma-many things that it, it used. So I, yeah, I'll point out some things that really stood out to me on the technical front. So, you know, the f-first thing I noticed was that in the benchmarks it scored, um, exceptionally well at this AALCR benchmark, which measures a model's ability to, um, to reason about and to synthesize information from long-form documents.

\[00:13:56\] **Ben Lloyd Pearson:** So we're talking like 10 to 100,000 tokens, uh, which is, \[00:14:00\] you know, that's a pretty notable achievement 'cause that is still a thing that a lot of models are, struggle with, particularly other local models. Uh, but then really what's notable is that the, you know, they, it only requires 20 gigabytes of memory to run, which is, um, you know, seems like it's a, it's a, like exceptionally low.

\[00:14:18\] **Ben Lloyd Pearson:** Um, but then there were some, uh, new developments in this that, that were also just in-interesting as like technological incremental improvements. Like, uh, they have this concept called speculative decoding, which is where they have this like super lightweight model that generates the output tokens in larger chunks rather than sequentially, which is how a lot of models do it today.

\[00:14:40\] **Ben Lloyd Pearson:** Um, and then those chunks are sort of validated after the fact by a smarter or bigger model, uh, to confirm their validity. So yeah, lots of just really cool, like incremental in-in-innovations out of this that, you know, are really making the local model i- space really seem like it's starting to heat up. And I think the coming months are gonna \[00:15:00\] be, like a lot of attention's gonna be not just on this, but on all of the developments happening to local models.

\[00:15:07\] **Andrew Zigler:** I agree

\[00:15:08\] **Ben Lloyd Pearson:** Yeah. And speaking of open source, let's talk a bit about why open source matters for AI. So we got an article here from the O'Reilly Substack from Tim O'Reilly bringing some just really great sage advice from, for, for the AI era from somebody who has been around for a lot of major technical or technological developments.

\[00:15:30\] **Ben Lloyd Pearson:** And this article argues, you know, how, you know, the open source models, the, the weightings, the harnesses, the context layers, the data that goes into this, it's really important that we do have, um, effective open source competitors in this space because, uh, you know, there's a lot of risk in relying too heavily on a small number of proprietary firms to provide these types of services.

\[00:15:55\] **Ben Lloyd Pearson:** So there's a really great, um, allegory in here to, you know, the \[00:16:00\] early days of the web when you had Netscape and Microsoft who were sort of duking it out and trying to figure out, like, how do we own the entire tech stack of the web? Like, can we own the servers and the, the client side and just try to monopolize all of it?

\[00:16:14\] **Ben Lloyd Pearson:** And, you know, for a while, it seemed like that may actually play out, but then you had something like Apache hit the scene, uh, that, you know, quickly followed with like the LAMP stack becoming the norm, and suddenly everything is open source and nobody fully owns the tech stack of the web. And, you know, some of the risks that O'Reilly highlights in this article that I think are really worth, you know, just paying attention to is that, you know, if we're, if we're increasingly relying on like one or two or three, uh, frontier labs, uh, to define like things like personality traits and the guardrails that go into these, um, you know, the, the, there's a risk of everything sort of like going towards the lowest common denominator.

\[00:16:56\] **Ben Lloyd Pearson:** You know, we, we all start to become the same with the same outputs and the same \[00:17:00\] approaches to solving problems. And, um, and there's a real risk that that sort of, um, stifles a lot of create, creative in- innovation, uh, for, for lack of a better phrase. Um, so, you know, when, when you think about how like engineering teams are tinkering with a lot of AI today, like if you're using the frontier stuff, you're really doing more ar- around customizing your harnesses rather than customizing like the weights of the model, for example.

\[00:17:29\] **Ben Lloyd Pearson:** And you may actually... Like both of those may be important things to focus on. So yeah, I just really like this as, you know, somebody who's, who's been in this industry for a very long time and has been successful and, uh, you know, and he's a big friend, O'Reilly's a huge friend of open source and long has been.

\[00:17:45\] **Ben Lloyd Pearson:** It's just really great to hear his perspective on how AI is shaping things. So what did you think about this article, Andrew?

\[00:17:52\] **Andrew Zigler:** I thought it was really smart how the article calls out that the, where the, seams of \[00:18:00\] uh, of open source technology is used for models. It's, up until now, there's been a lot of controversy around what open source even means for a model. Like, "Oh, you give us

\[00:18:09\] **Ben Lloyd Pearson:** Yeah

\[00:18:10\] **Andrew Zigler:** that's great, but did you give us the training code?

\[00:18:12\] **Andrew Zigler:** Did you give us the weights? Did you give us, uh, the corpus that was used to train it?" All of those things have their check boxes, and people use them to grade if something's open source. But in this article from O'Reilly, he's really focusing on how the thing we need to focus on making open source is the modularity of the stack that all of the AI stuff is operating on, the inference platforms, all of the tooling that we use to serve and store the data for them. Making them as composable and modular and open as possible is actually the keys for letting all of the rest of that thrive. Because to your exact point, um, like you, you get in this situation where different large players own really critical parts of just, like, the baseline experience of using the model.

\[00:18:59\] **Andrew Zigler:** Like, think \[00:19:00\] of what we've covered so far. We talked about Anthropic really becoming, uh, really having this, you know, a grip on the, on the classifying the, the dangerous commands and putting guardrails on prompting and stuff. Like, if you move into an open source world, you don't have that anymore. So we have to think about what are the parts that give us the equivalents, and that's where this really cool, like, AI potluck initiative that he calls out comes from.

\[00:19:22\] **Andrew Zigler:** The idea of like, how do you build this very rich ecosystem? Think like Linux Foundation level rich ecosystem of all of the parts you need to run a cloud and making it open source. We need that equivalent for AI, and that's what the potluck is. That's a pretty cool initiative, and we'll link it for folks to check out. Um, I think that this is a really important part that we get right, and so far I think we are. I think of the major tooling and the parts that I use to run my open source models or my harnesses and, um, you know, I, I feel like I have the parts I need to at least start assembling together. Uh, but that's where I think a lot of the challenges will live, is supporting \[00:20:00\] that ecosystem and preventing all of us from fracturing into just writing our own versions with these models.

\[00:20:07\] **Ben Lloyd Pearson:** Yeah. And, and there's, there's a lot in this article that I, I really, uh, strongly agree with. Uh, so I think, y- you know, all of our listeners should... It's, it's worth taking a, uh, giving it a read. Uh, but the tone in it kind of made me feel almost like, and we talked about this before recording, it, it was almost like a very cautionary and sobering take.

\[00:20:26\] **Ben Lloyd Pearson:** Like, it, it's like, you know, like a very just like, um, uh, you know, almost cold like viewpoint on this perspective. Like, I, I don't wanna say pessimism, but it, it was almost feeling like it was bordering on having a pessimistic take about it. Um, but I, you know, I think I really have sort of a very different perspective on it.

\[00:20:44\] **Ben Lloyd Pearson:** Like, you know, I think that, yeah, open source is being very heavily disrupted today because of AI, and in some ways it, it is, it is sort of following the lead of these proprietary companies when it comes to the frontier of AI rather than being what's leading \[00:21:00\] it. Um, but I also think that we're sort of primed for a bit of like an open source renaissance actually, because, you know, you mentioned distillation earlier with the, the, with Meta's new model.

\[00:21:11\] **Ben Lloyd Pearson:** Um, you know, AI has made it easier than ever to replicate and iterate on other people's ideas. And, you know, I think open source is just going to continue very closely tracing the capabilities of the frontier model companies. Uh, you know, just particularly consi- considering again how easy it is to distill value out of stuff that exists.

\[00:21:31\] **Ben Lloyd Pearson:** Uh, you know, we saw Claude, um, some of their, uh, code and architecture get released or leaked earlier this year, and immediately everyone's out there with like their own versions of how they've, you know, they've distilled it into something else. So, you know, it's a very, very exciting time in, in, you know...

\[00:21:47\] **Ben Lloyd Pearson:** But again, a great take from O'Reilly on the state of things. All right, let's talk now about your SDLC and context engineering. So here we have an article from \[00:22:00\] our friends over at Lead Dev, um, where it really just breaks down how, you know, there's so much context, uh, scattered across all of your SDLC that really needs to be pulled into your agentic systems for them to make good decisions.

\[00:22:15\] **Ben Lloyd Pearson:** So, you know, there's, there's a lot of context that exists in the way that work is specified and how it's reviewed, where, where and how it's tested and, you know, the realities of, of it being shipped. Um, and you really need to be accounting for all of that when you're trying to build an agentic system to contribute code into production.

\[00:22:36\] **Ben Lloyd Pearson:** So you need to think about stuff like documenting all the states of your life cycle. You know, what's the current state of your project? What's the future plans for your project? Um, what, what's permanent and should, should never change without like a great deal of, um, you know, focus. Um, and really understanding all of those things and giving them to your agents when they need it \[00:23:00\] within the SDDLC.

\[00:23:01\] **Ben Lloyd Pearson:** Um, and this article, you know, it kind of-- it's a very broad article. It covers a lot of topics, so it's kind of a little bit too much that we can cover in, uh, length here. Um, but you know, I, we, I wanted to bring it up because, you know, this con- this idea of a context layer for your SDLC is a topic that we're continuing to see come up like over and over again right now.

\[00:23:22\] **Ben Lloyd Pearson:** And when you really think about it, you know, early agents really just had access to like your code base. Maybe they had some basic metadata like your project management tasks or PR descriptions. But I think most of us learned like very early on that that wasn't enough context for most engineering decisions.

\[00:23:42\] **Ben Lloyd Pearson:** There were some places where it w- was enough, but many places where it was not. And, uh, you know, there's a lot more that goes into making good engineering decisions. So, you know, you have to think about, like, how well-planned is your spec? You know, what learnings and decisions did you make along the way that \[00:24:00\] modifies your final outputs?

\[00:24:02\] **Ben Lloyd Pearson:** Um, what parts of the code base have risky components that need particular care or review when modifying? Um, you know, which components get bogged down in reviews the most? Like, these are all questions that you have to weigh when you're initiating a project or task, uh, because they all impact, like, the feasibility of, of, uh, completing them.

\[00:24:22\] **Ben Lloyd Pearson:** And, you know, humans, you know, we accumulate answers to these through experience, but your agent only has as much experience as you feed into it. Like, we covered this concept a while back of where, um, AI agents are like tourists. They, the, the, it's like the first time they've ever-- Every time they show up, it's the first time they've ever been there, and they only have as much context as they deliver, as you deliver to them as they go along their journey.

\[00:24:44\] **Ben Lloyd Pearson:** And then once they're done, they leave forever and go back home. Um, but that's, you know, that's really, like, you know, and this is something that, you know, LinearB, like we've really have started to ingrain this into the, to what we're building for our customers. You know, uh, we're really trying to help \[00:25:00\] understand, like, how do you, how do you get that context layer into all of your, your agentic decisions?

\[00:25:07\] **Ben Lloyd Pearson:** You know, and, and that's whether it's humans or agents. You know, if a human is making decisions, they should have a format of all of this context that, uh, works for them as well. Uh, so yeah, we've, we've been doing this a lot with customers recently, and it's, and I just think it's a really fascinating, uh, concept that really every engineering leader needs to be solving right now.

\[00:25:24\] **Ben Lloyd Pearson:** So Andrew, what did you think about it?

\[00:25:27\] **Andrew Zigler:** Yeah, really great coverage on, on, on what this article gave us. It definitely dove into different parts of what matters for a team working on a code base together. I think that's a real big focus is of times in agentic coding, you get these two different groups that are talking to each other. You have, like, vibe coder or someone with a project that they are the only person touching anything on it, and they can go really fast, and they don't have these kinds of guardrails.

\[00:25:49\] **Andrew Zigler:** And then you have folks that are using it as a team and trying to build a product together, and that's like a team sport. And so, uh, like, they need to have a lot more context sharing. And a lot of the techniques that \[00:26:00\] one velocity would just totally, know, wreck another one. And so that nuance is really important.

\[00:26:07\] **Andrew Zigler:** And this one gives us that, that secondary path, the idea of, like, what does this look like on a team level? Um, so the article gives us some really, uh, great tactics here. I think your point about, you know, the agents are almost like tourists, and they come in and they do their work and they leave. A lot of this is around how do you turn those agents into more like of this code base?

\[00:26:27\] **Andrew Zigler:** Like, they are native to it, fluent in it. They are... They understand the parts that are needed to operate just, you know, as part of their, uh, their, their operations themselves. And that's what's achieved by having these kinds of layers of context as he's, he describes here. I will say I wanted to go, you know, like on, like, a fridge, you get, like, the word magnets you can rearrange.

\[00:26:49\] **Andrew Zigler:** I really wanted to do that with the title 'cause I'm like, "Your SDLC is your context engineering," is not saying what I want, what this article says to me, and it's really more about context engineering \[00:27:00\] for your SDLC or, you know, how to put a harness on your SDLC. Because that's really what this is unlocking is it gives, uh, you as, like, an operational or a leader person the vantage into this, uh, code world that's shared by a bunch of folks.

\[00:27:14\] **Andrew Zigler:** But then it also gives those folks and their agents the real time relational context about what the heck is moving around them. And I mean, gosh, we talk about that all day here on Dev Interrupted because that's our story at LinearB. So really resonated with this article.

\[00:27:32\] **Ben Lloyd Pearson:** Yeah. And speaking of context for your agents, let's talk about personalized skills and some new research that came across our desk on whether or not they help coding agents. So this is a new study that we read, uh, around the concept of personalized skills. Uh, and these are preferences that'll learn from an individual developer's past interactions with AI coding agents.

\[00:27:55\] **Ben Lloyd Pearson:** And it-- the research wanted to find out if those were better for, \[00:28:00\] um, AI coding agents than more generalized skills. So they ran this test, uh, across 13 developers with over 200 real-world developer agent sessions. So it's a little bit of a limited, uh, sample size, but enough to at least get some ideas about what might work.

\[00:28:17\] **Ben Lloyd Pearson:** And there's actually kind of some surprising findings. Um, y- while the personalized gains did, or the personalized skills did provide some productivity gains, um, it was actually, like, kind of marginal. Um, particularly when you looked at the impact of more generic skills that, um, are pooled across many developers.

\[00:28:37\] **Ben Lloyd Pearson:** Um, those are the ones that have, that produced the largest and the most consistent improvements. And, you know, there's-- w- we've-- I, I feel like there's been an ebb and a flow over the last two years or so where, um, people, you know, think that maybe AI needs to, you know, be heavily customized to the individual situation or person or, \[00:29:00\] or workflow.

\[00:29:01\] **Ben Lloyd Pearson:** Um, or does it need to be more broadly, you know, constrained to, to have like, um, wider practices? Um, and, you know, this research seems to indicate that particularly when you're, you're trying to look at like what's gonna have the biggest impact on an organization, it's probably better to have skills that, that help everyone a little bit, rather than to have one person-- help one person a lot in just a few situations.

\[00:29:26\] **Ben Lloyd Pearson:** So yeah, it's, uh, really great research to read for anyone that, you know, is out there buying or thinking about AI coding tools or building them, uh, as a way to sort of prioritize, like where should be, you would be investing your time for your team. So what did you think about this research, Andrew?

\[00:29:42\] **Andrew Zigler:** Uh, this was a really great, um, r- for me, a revisit back to when we had, um, Karthik Ramgopal, the distinguished engineer from LinkedIn, on the show talking about how to distribute gains across the entire engineering department. We've had a lot of, uh, big leaders on the show at, like, b- you \[00:30:00\] know, established enterprises and, like, not small engineering teams manage to get, um, these kinds of operational gains.

\[00:30:06\] **Andrew Zigler:** And then also get those gains across adjacent departments, marketing and finance and HR, and how it all comes down to is having this distribution system. A, a place where the experiments can live together, a place where learnings like memories and skills can live together, be version controlled and distributed, uh, amongst others.

\[00:30:29\] **Andrew Zigler:** And so that's the really big takeaway from this, is that, know, I, I felt like this article was coming for me a little bit 'cause I have lots of personalized skills based on my own

\[00:30:38\] **Ben Lloyd Pearson:** Yeah.

\[00:30:38\] **Andrew Zigler:** with all sorts of stuff, and frankly, I think a lot of, uh, like agentic operators do. Uh, you just kind of accrue them over time, and this article really calls out that maybe the benefits of those are marginal or not as much as you think. I argue maybe there's a compounding effect of using a lot of those together to create, do a very domain specific task, but maybe that argues that there's more for simplification to be \[00:31:00\] done, which I agree with. But the b- the biggest thing here is that those gains are marginal. If you distribute those same kinds of, uh, learnings, but on a general level and to everybody, the gains in productivity across the board are just substantial, substantially larger than they can be for the individual.

\[00:31:20\] **Andrew Zigler:** This really speaks to the power of us pulling together what works, especially within an organization. So, uh, and it's really promising data that, like, this is a real trend now captured in this research, so really great, uh, dive. I recommend, uh, folks check it out when we include it. Um, but if you haven't thought about how you're distributing skills within your team, whether a engineering team or a product team or otherwise, like I think that is your immediate next opportunity.

\[00:31:47\] **Ben Lloyd Pearson:** Absolutely. Absolutely. Well, Andrew, what are your agents up to this week?

\[00:31:52\] **Andrew Zigler:** Well, they're being cautious because I'm basically out of tokens, and they have

\[00:31:57\] **Ben Lloyd Pearson:** Ugh

\[00:31:57\] **Andrew Zigler:** concept of throttling themselves when that happens. So, \[00:32:00\] um, I, I'm getting caveman talk again, unfortunately. We're back in caveman days. Um, but, but, uh, uh, thankfully, I'm only a few hours away from a refresh, so then we'll be back. Um, I guess they'll be talking like Shakespeare or Homer or whatever they feel like. What about your

\[00:32:15\] **Ben Lloyd Pearson:** I

\[00:32:15\] **Andrew Zigler:** What are they up to?

\[00:32:16\] **Ben Lloyd Pearson:** feel like you need to get a little bit of sub-agent delegation and model routing into your life from the sounds of it.

\[00:32:22\] **Andrew Zigler:** just need to take some time to try out the new Glimmer and kick some things over to that. I will say I burned my tokens doing a lot of really cool stuff this week, like, uh, working on a lot of news reading stuff. I love to read lots of stuff, but as you know, the world's moving way too fast. So we talked about an article in here recently about, like, there's too much information to consume it all.

\[00:32:42\] **Andrew Zigler:** Like, how are we reading? How are we writing? And it talked

\[00:32:44\] **Ben Lloyd Pearson:** Yeah

\[00:32:45\] **Andrew Zigler:** like, layer of, of, like, understanding what's going around in the world and the curating what matters for you. So I've been trying to explore that and build that, know, using a lot of tokens in the process.

\[00:32:55\] **Ben Lloyd Pearson:** Yeah.

\[00:32:55\] **Andrew Zigler:** you?

\[00:32:55\] **Ben Lloyd Pearson:** we j- we just talked about generalized skills. I, I've, I-- You know, with \[00:33:00\] these Mythos class models now all out, you know, we got s- we got all of them, Sonnet, Opus, Fable. Uh, I can kind of pick and choose now. I've decided to kind of go back and, like, look at some of the stuff that, that we've built in the past and how it's, you know...

\[00:33:12\] **Ben Lloyd Pearson:** A lot of it hasn't, has-- It needs to be it- it needed to be iterated on to keep up with, uh, where the models are today. So, uh, yeah, definitely some, uh, like, generalized skills happening. But then also just, like, it's a good time to build. As you mentioned, it's a great time to build new skills because these Mythos class, uh, models are, you know, the Opus 5, Fable 5, they are really great at constructing these things.

\[00:33:37\] **Ben Lloyd Pearson:** So, um, yeah, mostly just trying to get toil out of the way, you know. That's r- really where we're at right now. It's like there's so many opportunities to just be like, "Hey, that takes too much time. I'm gonna go have AI do it for me now."

\[00:33:49\] **Andrew Zigler:** And every week it gets a little easier too

\[00:33:51\] **Ben Lloyd Pearson:** Yeah. Well, speaking of every week, thanks again to our listeners for joining us again this week for "The Friday Deploy," presented to you \[00:34:00\] by LinearB.

\[00:34:01\] **Ben Lloyd Pearson:** It's always a pleasure to get to share our, our opinions on and our learnings on what's happening in the space of AI and agentic development. So, thanks for joining us again today. Um, engage with us wherever you find us out on social media. We're out on LinkedIn, we're on Substack. You can see us on YouTube as well.

\[00:34:16\] **Ben Lloyd Pearson:** Leave a comment, a like, a thumbs up, whatever you can do to help us. It, it really does just help spread the word of the show. So thanks for sticking around to the end, and we'll see you next week.

\[00:34:26\] **Andrew Zigler:** See you next time

## Real conversations with top engineering leaders

Find us on

[](https://www.linkedin.com/showcase/dev-interrupted/)
[](https://devinterrupted.substack.com/)

## Your next listen

[![Cover image for The playbook to close your team’s AI productivity gap | LinearB’s Yishai Beeri](https://assets.linearb.io/image/upload/c_limit,w_3840/f_auto/q_auto/v1/Blog_Comprehensive_DORA_Guide_2400x1256_74_0278e11de6?_a=BAVMn6ID0)](https://linearb.io/dev-interrupted/podcast/linearb-yishai-beeri-2026-ai-productivity-gap-pr-yield)

Dev Interrupted

[The playbook to close your team’s AI productivity gap | LinearB’s Yishai Beeri](https://linearb.io/dev-interrupted/podcast/linearb-yishai-beeri-2026-ai-productivity-gap-pr-yield)

LinearB CTO Yishai Beeri joins the show to unpack 2026 benchmark data from 2.7 million PRs, exposing a widening productivity gap across engineering teams....

[![Cover image for Model welfare, building a civilization for agents, and the CI/CD landrush](https://assets.linearb.io/image/upload/c_limit,w_3840/f_auto/q_auto/v1/_797274efda?_a=BAVMn6ID0)](https://linearb.io/dev-interrupted/podcast/steve-yegge-agentic-civilizations-ai-harnesses-cicd)

Dev Interrupted

[Model welfare, building a civilization for agents, and the CI/CD landrush](https://linearb.io/dev-interrupted/podcast/steve-yegge-agentic-civilizations-ai-harnesses-cicd)

On this episode of the Friday Deploy, Ben and Andrew break down Steve Yegge's radical approach to building agentic civilizations and pushing code straight to...

[![Cover image for Why AI gains are unevenly distributed in your engineering team | Asana’s Arnab Bose](https://assets.linearb.io/image/upload/c_limit,w_3840/f_auto/q_auto/v1/Blog_Comprehensive_DORA_Guide_2400x1256_72_a71bae404c?_a=BAVMn6ID0)](https://linearb.io/dev-interrupted/podcast/asana-arnab-bose-ai-productivity-agentic-work-management)

Dev Interrupted

[Why AI gains are unevenly distributed in your engineering team | Asana’s Arnab Bose](https://linearb.io/dev-interrupted/podcast/asana-arnab-bose-ai-productivity-agentic-work-management)

Asana Chief Product Officer Arnab Bose explains why enterprise AI gains remain unevenly distributed across engineering teams. Discover how moving from isolated...

## Structured data

_Machine-readable metadata (JSON-LD) embedded in the page for search/AI context — not content rendered on the page itself._

```json
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "LinearB",
  "url": "https://linearb.io/",
  "logo": "https://assets.linearb.io/image/upload/v1715628027/logo-mark-lg.svg",
  "description": "LinearB is the engineering productivity platform that helps engineering leaders prove AI is improving throughput without sacrificing delivery confidence, flow efficiency, or developer experience.",
  "sameAs": [
    "https://www.linkedin.com/company/linearb"
  ],
  "award": [
    {
      "@type": "Award",
      "name": "LinearB is a Leader in the 2026 Gartner® Magic Quadrant™ for Developer Productivity Insight Platforms",
      "dateAwarded": "2026",
      "awardedBy": {
        "@type": "Organization",
        "name": "Gartner®"
      }
    },
    {
      "@type": "Award",
      "name": "Great Place to Work Certification",
      "dateAwarded": "2025-2027",
      "awardedBy": {
        "@type": "Organization",
        "name": "Great Place to Work"
      }
    },
    {
      "@type": "Award",
      "name": "America's Best Startup Employers 2025",
      "dateAwarded": "2025",
      "awardedBy": {
        "@type": "Organization",
        "name": "Forbes Magazine"
      }
    }
  ],
  "hasCertification": [
    {
      "@type": "Certification",
      "name": "SOC 1 Type 2"
    },
    {
      "@type": "Certification",
      "name": "SOC 2 Type 2"
    },
    {
      "@type": "Certification",
      "name": "GDPR Compliance certification"
    },
    {
      "@type": "Certification",
      "name": "ISO 27001"
    }
  ]
}
```

## More on linearb.io

### Top navigation

- [Book a Demo](https://linearb.io/book-a-demo)
- [AI Code Reviews — Catch security risks, bugs, and spec mismatches](https://linearb.io/platform/ai-code-reviews)
- [AI & Productivity Insights — See how AI tools affect cycle time and delivery speed](https://linearb.io/platform/ai-developer-productivity-insights)
- [Measure AI Impact — Track AI adoption and tie it to delivery outcomes](https://linearb.io/use-case/measure-ai-impact)
- [MCP Server — Chat with your data to spot patterns and boost output](https://linearb.io/platform/mcp-server)
- [Resource Allocation — Cost initiatives and shape your investment strategy](https://linearb.io/platform/resource-allocation)
- [Cost Capitalization — Capitalize engineering costs with audit-ready reports](https://linearb.io/platform/cost-capitalization)
- [Dev Team Management — Set targets and tie throughput to business outcomes](https://linearb.io/platform/goals-and-reporting)
- [DevOps Workflow Automation — Policy-based PR routing, approvals, and tests](https://linearb.io/platform/ai-workflow-governance)
- [AI Powered Support — Unify AI and human code delivery in one clear view](https://linearb.io/use-case/ai-powered-support)
- [Optimization — Surface friction with feedback and MCP insights](https://linearb.io/platform/developer-experience)
- [Reporting — Spot what's working and what needs attention](https://linearb.io/use-case/measuring-developer-experience)
- [Surveys — Turn developer feedback into actionable signals](https://linearb.io/platform/developer-surveys)
- [Platform overview](https://linearb.io/platform/overview)
- [Register now](https://linearb.io/event/the-great-software-factory-debate-2)
- [Customers](https://linearb.io/customers)
- [Pricing](https://linearb.io/pricing)
- [Why choose LinearB — Explore your data. Measure performance. Act to improve it.](https://linearb.io/why-linearb)
- [APEX framework — The operating model for AI-era engineering teams](https://linearb.io/resources/apex-framework)
- [Anti-FAQ — The questions other vendors won't answer](https://linearb.io/why-linearb/anti-faq)
- [Security — Enterprise-grade compliance and zero code access](https://linearb.io/security)
- [Build vs. buy — The hidden cost of building it yourself](https://linearb.io/resources/build-vs-buy)
- [Dev Interrupted Podcast — Conversations with engineering leaders](https://linearb.io/dev-interrupted/podcasts)
- [Reports & Guides — Deep dives on productivity and delivery](https://linearb.io/resources)
- [Webinars — Expert sessions on productivity and AI](https://linearb.io/resources?category=workshops)
- [Metrics Benchmarks — See how your engineering org stacks up](https://linearb.io/resources/software-engineering-benchmarks-report)
- [Blog — Product updates and practical insights](https://linearb.io/blog)
- [Help Center — Documentation, setup, and support](https://linearb.helpdocs.io)
- [API Docs](https://docs.linearb.io/api-overview)
- [Status](https://www.linearbstatus.com/)
- [Integrations](https://linearb.io/integrations)
- [LinearB Library](https://linearb.io/library)
- [Engineering metrics](https://linearb.io/library/engineering-metrics)
- [Platform engineering](https://linearb.io/library/platform-engineering)
- [Engineering glossary](https://linearb.io/library/engineering-glossary)
- [Developer productivity](https://linearb.io/library/developer-productivity)
- [AI in software development](https://linearb.io/library/ai-in-software-development)
- [Engineering management](https://linearb.io/library/engineering-management)
- [Developer experience](https://linearb.io/library/developer-experience)
- [DevOps](https://linearb.io/library/devops)
- [Engineering operations and the context layer](https://linearb.io/library/engineering-operations)
- [Engineering efficiency](https://linearb.io/library/engineering-efficiency)
- [Software delivery](https://linearb.io/library/software-delivery)
- [Research and data](https://linearb.io/library/engineering-benchmarks-and-research)
- [LinearB is a Leader in the 2026 Gartner® Magic Quadrant™ for Developer Productivity Insight Platforms](https://linearb.io/resources/gartner-magic-quadrant-dpi-platforms-2026)
- [Sign in](https://app.linearb.io/login)
- [Enterprise](https://linearb.io/solutions/enterprise)
- [Contact](https://linearb.io/contact-us)
- [About us](https://linearb.io/about-us)
- [Careers](https://linearb.io/careers)
- [Service agreement](https://linearb.io/services-agreement)
- [Privacy policy](https://linearb.io/privacy-policy)
- [DPA](https://linearb.io/data-processing-agreement)
- [Security FAQ](https://linearb.io/security-faq)
- [Substack](https://devinterrupted.substack.com/)

### Footer

_Additional links from the site footer, not repeated from the top navigation above._

- [GitHub](https://github.com/linear-b)
- [LinkedIn](https://www.linkedin.com/company/linearb)
- [Twitter](https://twitter.com/LinearB_Inc)