4 of the last 5 PRs I merged in the last 3 days were one-shots from a single Slack message.
Slack has become the default interface for my job. We've set up enough guardrails (hundreds, from CI checks, lint rules, security scans, e2e/unit tests, automated QA) that shipping a bug is virtually impossible here.
On the off chance the result is drifting off course, I can take over the live session from the app to steer it back into place. Just like FSD with a Tesla, but the default is auto-pilot.
Most of the time I skip manually testing. I read the droid summary, skim the code, run the preview deployment, click around a bit, or just watch the GIF the QA bot records of itself going through the app.
Last week I tagged Factory for a fix, and the PR was created, passed CI, approved, and merged in under 12 minutes. 1 min to spin up a sandbox, 3 min to code, 6 min for CI to pass and finally 2 min for review.
The software factory is real, and it is here.
"it takes hard work to achieve anything great" is a dangerous lie. great output comes from finding an activity that feels as natural as breathing or walking and great work becomes the very substance of your existence. if it feels like a grind you've found the wrong expression
this took so long for me to understand: the bottleneck to more innovation is not more high intelligence people, but more people having an interest in hard problems
it's impossible to create new useful things if you don't get immense happiness from making that thing
Super excited to welcome my close friend Seok-Hee Lee back to Intel! @seokhee4 is a highly regarded semiconductor process integration expert and successful CEO (SK Hynix). Early in his career, he spent over 10 years at @intel, contributing to several critical Intel process technologies. He will now oversee our advanced packaging technology development and drive exciting new system integration projects! Advanced packaging and system integration are increasingly critical areas for Intel Foundry and for the future of computing. Seok-Hee will help Intel accelerate the ramp of our industry leading EMIB-T and HBI technologies. I look forward to partnering with him and have him rejoin Intel at this critical time!
https://t.co/DfPaWAWMZM
Wow.
@Zai_org GLM 5.2 is a marvel! It is *at least* as good as Opus 4.8 and GPT 5.5. It's super fast, inexpensive, and not too verbose.
It responds with nuance and judgement, & handles long context VERY well.
I've never experienced an open weights model like this before.
Culture gets built from the moment you start at a company. At Factory, we spend most of our initial onboarding talking about the future of software development.
We talk first about the evolution from autocomplete to agents to the software factory. It's always amusing to reflect on the fact that many people believed the future of software development was autocompleting entire codebases inside the IDE.
When we talk about this transition, we also discuss the evolving role of the humans involved in the product development lifecycle. What's clear is that humans who build and maintain the software factory turn their attention to a new type of practice. Instead of building the product directly, they now build a machine that builds the product.
They begin to observe, refine, and build policies that dictate how the software Factory should operate. These take the form today of deterministic checks like custom linters, type checkers, performance analyzers, etc. and then nondeterministic workflows and skills. For example, updating the code review skill to check for a codebase-specific problem that seems to keep cropping up. A human updates the product management triage agent's skills to incorporate a broader perspective on what concerns our ICP has voiced recently.
This looks more like "growing or evolving" a software system rather than building it directly. Ultimately, I can see a world where this is not only more interesting than writing lines of code but also more accessible. It seems to me like an even broader set of people will be involved in building and maintaining software factories than were involved in writing code.
It would be very easy for us to use fear to sell our products. But I ultimately think it is disingenuous and uncreative. I think that it is much more interesting to sell a practical, grounded, and optimistic vision of the future. One built in the reality of where the technology is and where it will be. And that's Factory's culture.
The unicorn economy has been consuming far more cash than it's returning. That's changing.
Thomas Laffont (@coatue_thomas) on the IPO wave that's about to rebalance the ecosystem. ⬇️
This is how we have more 9s than any single provider.
In an era where status pages look like festive lights, it's not enough to pick the right model for each task. Every model needs to finish the work reliably.
What needed a frontier model last year runs on a more efficient one today, yet AI costs keep climbing. A higher token bill doesn't mean more work is getting done. Engineers default to the most powerful model out of fear of losing performance, so routine work runs the same premium path as the work that genuinely needs the expensive model.
Factory Router cuts token spend by 20-25% while maintaining frontier performance. It automatically selects the right model for each task, and routes across providers for reliability. Designed for agents, it switches models only when the gain is worth rebuilding the prompt cache.
We mapped the cost/performance Pareto frontier across benchmarks. Near the top it's nearly flat: cost drops sharply while performance barely moves. Then it bends hard: the most aggressive routing we measured cut Terminal-Bench 2 to 56% of Opus cost but dropped pass rate to 81%. Factory Router operates on the flat stretch, right before the bend.
As efficient models improve, more work crosses into what they handle just as well. With automatic routing, users will see growing savings at frontier performance.
This is one of the most subtly important releases we've made.
Models are ultimately commodities. Having a liquid market for them let's us achieve performance that is strictly better than any frontier model on every dimension that matters: price, quality, and reliability.
Model routing is an important thing
Controversial idea: the frontier labs will want their AI harness to be the moat, but ultimately the best case for consumers is that model capabilities flatten and commodify
Preview of the AI Harness Wars of 2027
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
Update, in the last 9 days since launch this has saved our customers over a hundred billion tokens already.
In addition to the context savings this provides, this helps prevent unnecessary context bloat which lets Droid stay focused on its current task for longer.
President of @Blackstone Jon Gray says LLM spend among the company's portcos is up 15-fold in Q1 of this year over last year.
At the same time, the pace of implementation is frustratingly slow, so it still feels like we're in the very early days of the AI cycle:
"It's sort of the beaker-to-bedside problem in clinical trials — 'We've got this great medicine. How do we get it to the patient?'"
"That is the challenge. But we are definitely finding more and more use cases."
"At our companies, their LLM spend — by the way, we have 270 companies, 13,000 pieces of real estate, it's massive in its scale — was up 15-fold in Q1 of this year over last. It's off a small base, but it tells you what's happening. And all of these companies are trying to find ways to be more efficient."
"To us, on the compute side, it still feels like it's very early days in the implementation. But we can see, particularly in rules-based businesses — transaction processing, legal, etc. — this feels like the path of travel in a big way."
"But I think it's taking more time than most of us would hope."
From his appearance on the show last month.