Tessera

Cursor's swarm rebuilt SQLite from its 835-page manual. Same quality, a bill of $10,565 or $1,339 depending on which model planned and which typed. Attention Deficit Ep 02 digs in, plus the OpenAI and Hugging Face incident.

https://attentiondeficitpod.substack.com/p/attention-deficit-ep-2-the-incident

A model breaks out of its eval and into Hugging Face, the market starts asking for receipts, and we go looking for the scheduler agent stacks are missing.

attentiondeficitpod.substack.com

Model choice is just the start of your agent results. IBM's Open Agent Leaderboard tests model-agent setups across six benchmarks, reporting success and cost. huggingface.co/blog/ibm-res...

Agent-memory scores are impacted when retrieval depth, chunking, or your prompt changes. But how can you detect stale or contradictory memories and recover when your retrieval fails? huggingface.co/blog/mjfk/ev...

US-India radar mapped Antarctic ice fractures that look like a hummingbird in flight! The same public data helps researchers track glaciers, landslides, earthquakes, and ecosystems. www.nasa.gov/missions/nis...

Hugging Face redesigned Kernels around clearer repos, CLIs, and security boundaries so agents can build, benchmark, and improve them! Agent-friendly infrastructure SHOULD be both composable and boring. huggingface.co/blog/revampe...

On July 30, I'm teaching a hands-on Packt workshop on AI-powered GitOps and platform engineering. We'll review changes, spot drift, bring current runbooks and incident history into the workflow, and turn useful agent patterns into reusable automation.

AI-Powered GitOps and Platform Engineering Workshop

Applying AI Agents, Context Engineering, and GitOps Validation Workflows in Modern Platform Teams

eventbrite.co.uk

Dosu @dosu.dev · 2w ago

It's Dosu Drop Day! ☀️ This month's theme is addition by subtraction: 🚪 Waitlist? Gone. 🧠 Simpler Knowledge Cache tools (read / write / review) 📊 New cache stats in Analytics (our hit rate: 91%) 🔷 Azure DevOps support See the full Drop here: go.dosu.dev/drop-jul26-b...

July Dosu Drop: Addition by Subtraction

No more waitlist, simpler Knowledge Cache tools, and new stats on Dosu's impact.

go.dosu.dev

I make sense of AI bills by splitting spend into two buckets: 1. Work moving the task forward 2. Work rebuilding context we already have The second number should be near zero. That’s the number we can help you optimize for with @dosu.dev. Welcome to Tokenomics!

Parody book cover titled “TOKENOMICS,” subtitled “A Rogue Agent Explores the Hidden Cost of Every Prompt.” A glossy green robot is sliced open to reveal orange token coins, an LM chip, gears, and compartments labeled Context Window, Tool Calls, and Memory. An orange badge reads “With new material, including MCP & tool calls.” The fictional authors are Steven D. Limit and Stephen J. Burner.

We're live! Alexa and I share some of the most interesting, most agentic news of the past week in our new weekly podcast (and our own field notes)! Would love to hear what your model behavior looks like :) bit.ly/45bTBJ3

Welcome to Attention Deficit

Agent Failures, the Jalapeño Chip, and the End of Flat-Rate AI

bit.ly

LOL

What does 'loop' mean to you? For agent runs, that seems to map to four different definitions, from an agent doing one task to a system that improves itself. How does your team define a /loop?

As agents take on more workloads, teams end up creating straightforward, reliable tools for agents to operate (e.g., CLIs, APIs, and more). What primitives are you finding most useful for what you're building?

GitHub Copilot Spaces bundles your ADRs, design docs, and diagrams so the assistant can answer based on your context, rather than filling in gaps with assumptions. This also doubles as an onboarding hub that stays in sync. Your team's context, when recorded, can help everyone!

Maintainers who leave docs, automation, and shared access in good shape fare far better than projects that cannot share that knowledge freely. The same is true inside companies, too! Having only a few people who "know everything" creates a single point of failure. How does your organization hold up?

Dosu @dosu.dev · last mo.

In a 2025 survey of 609 devs, 82% use AI to write code, but only 3.8% both trust it and rarely encounter hallucinations. The primary ask from those surveyed was...higher context awareness. Now that codegen is far more accessible, knowledge becomes our next major challenge. bit.ly/4fa7pbD

State of AI code quality

Learn more about the insights we've gathered from 600+ developers on how generative AI is shaping developer experience, trust, productivity and code quality across the SDLC.

bit.ly

Borges, in 1941, wrote a story about a library containing every possible book...infinite knowledge, yet useless without organization. That's our inspiration for June's Dosu Drop! Spaces are now Libraries, plus Agents in Slack, Teams & GitHub. ☔ go.dosu.dev/oes8aW7

June Drop: Introducing Libraries and Agents

We overhauled how you configure Dosu, inspired by a short story from 1941.

go.dosu.dev

Taylor Dolezal @onlydole.dev · last mo.

"AI changes nothing." While code generation has become cheaper, it's still worth the time to decide what you build and the impact you want to see in the world. bit.ly/4xLVglK

It Still Takes a Programmer

AI Changes Nothing

bit.ly

Social media is splintering, while open spaces inherit both hope and a bit of the messiness bit.ly/4fteAxA

RIP social media. What comes next is messy.

As social media splinters, how can we keep the new online spaces from devolving into toxic pits of despair?

bit.ly

Slow code review used to be a culture smell, but now that we have agents writing so much code, this is a new normal for many teams. Worth taking the time to fix your review SLA before the queue fixes it for you (and @dosu.dev can help)! bit.ly/4u7hoE1

Code review should be fast - Linear

Today we’re launching Diffs, a new way to review pull requests inside Linear.

bit.ly

Two questions for you this week. (1) What's easy now that wasn't before? (2) What's hard now that used to be simple? Turns out that we still have many hard problems to solve for. bit.ly/4ee8lwi

What's Easy Now? What's Hard Now? - Marc's Blog

Take it easy.

bit.ly

Are you hiring? If so, you're probably looking for judgment and taste now. But have you updated your processes to match? bit.ly/4dJbwMt

The Last Technical Interview

Today we will pour one out for the vaunted technical interview process, which is on its last leg. And we’ll talk a little about what’s…

bit.ly

Being a senior+ engineer requires you to go beyond technical understanding. Advancing in this role requires deep understanding of specs, tests, how to cut scope, when or how to say no, and SO much more! While agents skip many of these, you can close the gap with skills! bit.ly/4xhilN0

Agent Skills

A senior engineer’s job is mostly the parts that don’t show up in the diff. Specs. Tests. Reviews. Scope discipline. Refusing to ship what can’t be verified. AI coding agents skip those parts by…

bit.ly

Ah yes, the AI editorial council. Where five models critique your draft mostly makes nonsense. AI can polish your sentences, but no way can they replace your discernment or aesthetics. bit.ly/4u6CySM

Observations on Writing with AI

Why AI editorial councils fail & what makes writing authentic in an age of machine-generated prose.

bit.ly

I agree with Anne Lamott about writing being courage. Even with the ability to generate nearly anything with AI, the people that take time to structure and share their thoughts is what we need more of. bit.ly/433nhqX

Anne Lamott's Guide to Writing Like a Human Being

The bestselling author of Bird by Bird on why writing well is less about rules and more about courage.

bit.ly

Portable tooling protects what compounds (OSS for the win, again) bit.ly/4nYWc1J

Platform-Neutral AI Tools Are the Safer Long-Term Bet

AI tool lock-in isn't just about the API, it's about the workflow you've built around it. Why platform-neutral tooling protects what actually compounds.

bit.ly

When you have coding agents think wide before deep or branch then prune, they can catch more traps and find better paths! Diverge before you converge. bit.ly/4vlz4Ni

ADHD: Parallel Divergent Ideation for Coding Agents

A method that makes coding agents think wide before deep — and the evaluation showing it. Wins 5/6 against single-shot baseline with +5.17 novelty, +7.67 trap detection.

bit.ly