Julie Sommerville

💡 AI doesn’t just change how code is written – it changes expectations and outcomes. In the final section of Jellyfish’s AI Adoption Guide, we explore how expectations for engineering leaders are evolving as AI becomes embedded across the SDLC. Because the organizations pulling ahead are treating AI as a transformation, not a tooling upgrade. Inside, you’ll explore: - Why AI adoption is quickly becoming a leadership mandate, not an option - How to balance ambitious goals with realistic expectations - What cultural investments are required to sustain long-term gains - How data can be used to align executives and engineering teams The teams that win in the AI era won’t just have better tools – they’ll have leaders who know how to guide change. Download today: https://lnkd.in/dut_6WpQ

20

2026 - *Clarity* is the scarce resource. Over the holidays, I've been thinking deeply about what are the real constraints w/ current versions of LLM assisted development? Watching Mathew Berman's latest video https://lnkd.in/eC7McJAU among other research, it is clear that the scaffolding is starting to catch up to the models, i.e. agents, workflows, CLI/IDE integration, AI code reviews, test generation, and longer running automation that turns “helpful tooling” into real shipping capacity. I think that the natural outcome of this is that implementation gets cheaper and faster, but *clarity* becomes the scarce resource. The real shift in 2026 will not be limited by writing code but more by defining constraints that keep outputs correct, safe, and aligned with intent: specs, invariants, security/privacy boundaries, and what “good” looks like.

34

1 Comment

How can you make sure orchestrating agents across many frameworks doesn't cost you a fortune? Today, Caelum Forder and I look into agent sessions from Coral Protocol.

16

Vibe: A Framework for Agentic Software Delivery I wrote up how Vibe approaches agentic delivery with a plan-first loop, tiered workflows (/vibe, /lite, /full), and “guards” as safety contracts that every build must pass before it can close. If you’re experimenting with agentic delivery and want something more structured than ad hoc prompting, this might be useful. Read the essay: https://lnkd.in/gA8in4SN

40

1 Comment

Voice agents and audio-native models are fast, fluid, realistic, and end-to-end; but every provider has their own messy, idiosyncratic API. If you've ever tried to use them yourself, you know how frustrating it can be.... Thankfully, spoken cleans things up! spoken is a unified abstraction over realtime speech-to-speech foundation models. Run any S2S model from OpenAI, Google, Amazon — one interface with one line of code. Try it out today! > pip install spoken > https://lnkd.in/eVvGHwge

85

7 Comments

Chunking and snippet extraction has been a huge focus lately - my latest blog dives into some of the work we've done on this to date, including support for a chunk rescorer in our semantic reranking retriever, as well as some useful ES|QL primitives to get more visibility into chunks and snippets. #elasticsearch #semanticreranking #snippets #chunks #chunking #esql https://lnkd.in/e6UD7iii

13

Next week I’m speaking at LeadDev LDX3 on the Ways of Working stage — all about team operating systems: what they are, why they matter, and how to build a great one! Hope to see you there. And to get in the zone, here’s a quick primer on systems thinking more generally. Huge thanks to GC.NYC #girlcrew - Woman-Owned Video Production — a phenomenal woman-owned video production studio — for bringing this video to life. Emily Geraghty and her crew are magic. If you’ve got an idea, reach out for help making it real!

82

10 Comments

New post from me on the complexity and importance of metric backfill in mobile observability. Backfill is typically an afterthought in traditional server-centric TSDBs, and it is anything but in the mobile world. Enjoy! (Blog link in the comments.)

30

1 Comment

💡 Heard a great concept on the Lenny’s Podcast recently: “Maximal acceleration” — the idea that the fastest way to go faster… is to start now. The key question? 👉 "How can we do it now?" Not next week. Not when it’s perfect. Now. It’s uncomfortable for most of us — but also incredibly freeing. Great companies are built on great execution culture. It hit home because this mindset is core at ICEYE. We obsess over unlocking learning and solving real problems now, not some other day. Even in a company that builds satellites — we strive to ask the question what would it take to do this today? Of course, most of the time we can’t actually do it today, but we build the right muscle memory for everyone. Now is the right speed.

41

2 Comments

This is a really interesting article about writing agents with the Claude Agent SDK. It gets into design advice that you should consider when writing an agent with Claude. It chimes with what I’ve read about the development of Claude Code (write as little business logic as possible and concentrate on getting out of the way of the model). The general takeaway for me is from the three-block flow chart in the middle: Give the agent good tools/affordances in three categories: Gather Context, Take Action, Verify work. Do that at the right granularity, with good feedback, and then let the model do the driving. It’s deceptively simple, but it has a goldilocks quality to it. Trying to make a more complex flow chart with many more boxes and lines constrains the AI to just the things you could think of in advance and makes it overfit to what you know today. Trying to make it simpler by, for example, giving the agent a whole existing API as a single tool can overwhelm it with too much choice and then too much context in the responses. It’s definitely worth reading if you're interested in building or using agents, irrespective of AI vendor. https://lnkd.in/eKF2rXWb

7

We’re excited to release Workflows 1.0 💫 - a standalone, lightweight orchestration layer for building multi-agent systems. It’s built on a few principles: 1️⃣ A lot of existing agent approaches are too opinionated (e.g. out of the box agent with tools) or too cumbersome to learn. 2️⃣ We strongly believe that an async-first, event-driven architecture is the right way to build agent systems, as opposed to something a bit more graph-based. The control flow can be defined in raw Python, and you can get all the efficiency benefits of async, while still ensuring there’s deterministic checkpointing. 3️⃣ It comes with the minimal, table-stakes things that helps any agent builder: human-in-the-loop, checkpointing, state, observability, composability, and more. It’s extremely lightweight and now it’s decoupled from our core LlamaIndex framework, so it’s super easy to install and use. Come give it a try! Huge shoutout to Massimiliano Pippi, Logan Markewich, and Zeyu Yang for working on this. Blog: https://lnkd.in/gA2kt-jW Python: https://lnkd.in/gmqU8dc4 Typescript: https://lnkd.in/g8bkfCh5

278

8 Comments

Claude Cowork = Claude Code for non-technical people I’ve watched smart leaders get excited by “AI”… then hit the wall the moment the work lives in folders, files, and formats. Claude Code was already a monster. I have been using as code agent and general agent. But it still required a terminal or techy interface and that’s a tax most non-tech people won’t pay. Cowork removes the intimidation layer and keeps the power: you hand Claude a folder, describe an outcome, and it plans + executes the work like a teammate. This is why I think it’s the most powerful tool release so far: it turns “agents” from a demo into a workflow normal people can actually use. Simon Willison pointed it at his blog drafts and it churned through 46 drafts, ran 44 site searches, and delivered a prioritized publish list. That’s not chat. That’s delegation. One sober note: it can edit and delete files, and prompt injection is a real risk so use a dedicated working folder and keep backups. If you’re rolling out GenAI in an enterprise, start a small Cowork pilot in a sandbox folder and share the wins (and the scars). https://lnkd.in/eEUEFaQC

16

4 Comments

"APIs are the connective tissue that allow an organization to thrive." Nearly every company I speak with is navigating the same challenge: modernizing their API landscape while preparing for AI-driven systems. The biggest brands want clear guidance on how to govern, scale, and future-proof their APIs as AI becomes more central to how software is built and operated. This conversation with Matthew Gray, Head of Channel Sales at Postman, does a great job exploring what that journey looks like in practice and why APIs sit at the center of it. Full discussion: https://lnkd.in/g8qUX9dt

126

5 Comments