Corey (CJ) Gwin (@corey_gwin) header
Corey (CJ) Gwin (@corey_gwin) profile picture

product dude. decade in startups. solutions to the right problems. product at https://teapawt.net/t.co/ltNOc6NdTV + https://teapawt.net/t.co/0NEfAW8E4J. this is my notepad.

Joined May 2008
Photo by Corey (CJ) Gwin (@corey_gwin)Photo by Corey (CJ) Gwin (@corey_gwin)Photo by Corey (CJ) Gwin (@corey_gwin)
All at once.
1
1
8
2,498
Corey (CJ) Gwin retweeted
my vocal synth models the vocal tract as a network of branching acoustic tubes, unlike vocaloid which stitches together and pitch-shifts phoneme recordings. still working on improving naturalness by tuning it against recordings of real singers, but I'm already impressed at how cool it sounds!
I only make serious software
211
603
84
5,708
435,303
Corey (CJ) Gwin retweeted
CANCEL your weekend plans. The 2027 meta just dropped. You NEED to: • Build a computer-use agent that parses the DOM, diffs screenshots and maps clicks to coordinates so it can drive legacy software with zero APIs • Ship generative UI where your LLM returns live React components instead of markdown blobs and watch your product feel 10 years newer • Wire a System-One router that resolves 95% of requests in milliseconds for fractions of a cent, escalating only the hard 5% to reasoning models • Kill your `while True` loop and rebuild it as a Temporal state machine that resumes from the exact node after a crash or a 3-day approval • Write a raw MCP server in pure JSON-RPC so any external agent can call your product with scoped credentials, no SDK required • Add grammar-constrained decoding so your agent emits 100% valid JSON on the first try, every try, forever • Build a trajectory grader in CI that blocks the merge the second a tool-call sequence deviates from the golden path • Ship an inter-agent sanitization proxy that executes prompt injections before they cross a single agent boundary • Wire a PII unmasking proxy that swaps sensitive data for UUIDs before inference and restores it after with full audit logs • Add a cost kill-switch that halts any loop projected to exceed $0.50 and degrades frontier → mid-tier → local → cache • Implement KV-cache prefix sharing at the gateway and cut Time-To-First-Token by 80% on repeated system prompts • Set up speculative decoding: local 3B drafts, frontier 70B verifies, acceptance rate logged on every single request • Shadow-route 5% of production traffic to your new model, diff trajectories and cost silently, auto-post the report to Slack • Automate the DPO flywheel: thumbs-down → preference pairs → nightly LoRA → eval-verified promotion, zero human touch • Build a voice loop with VAD and sub-1s interruption handling, streaming tokens and tool calls simultaneously • Ship an API-to-browser fallback that spawns a headless browser the moment a REST endpoint 404s, finishing the job anyway • Add a WebGPU shim so a 2B model runs in the browser when the cloud dies and your users never see a spinner • Fuzz your own agent swarm with malformed tool outputs and context overflow until it self-heals without a single pager alert • Publish one benchmark: your system vs the naive baseline, on cost, latency and task success, fully reproducible The wrapper era is over. The systems era pays. You have way too much to do. Bookmark & Repost.
49
99
14
1,506
154,191
Corey (CJ) Gwin retweeted
agents are going to cause a breakdown of human behavior assumptions underlying today's systems. i've started thinking about it as friction collapse. without friction, a number of norms will collapse. consider selling tickets to a concert/event. today, you can post them at a certain time and it's first come first serve. soon everyone's agent will persistently monitor, and as soon as they are posted for sale, thousands of people will instantaneously be trying to buy the same ticket. how do you even resolve this? sell the ticket to the person whose agent gets there fractions of a second before? we're seeing the same thing with job applications, restaurant reservations, but basically anything that relies on participation requiring effort is going to break down. sales emails/calls, participation in government benefits, apartment rentals, doctors appointments, RFP responses. customer support won't be able to handle the user who will not drop their point of dissatisfaction. with no friction, these systems just don't work anymore. business models will break down: - flights/hotels will be consistently cancelled and rebooked at the lowest price (great use case for your agent if you don't have this already set up) - subscriptions will be cancelled - retailers won't be able to arbitrage price differences as the agent will find the cheapest place to buy anything - insurance denials (health/auto/home/etc.) will be fought tooth and nail - bank interest rates will be optimized and avoidable charges will be avoided a lot of what we perceive as stable social coordination is actually an equilibrium sustained by human inefficiency. rationing is a product of human slowness. signaling effort/interest/intent is a key part of our social infrastructure. inertia is a key to many business models. information asymmetry is everywhere. and attention is monetized thoroughly. with agents, none of these things are true. humans will simply demand what they want and have their agents get it for them. there is going to be a fundamental renegotiation of how markets allocate resources, how people signal intent, and how businesses capture value.
177
232
89
2,122
198,422
my spidey senses are telling me the labs are, as we speak, using unguardrailed versions of their unreleased models to systematically reverse-engineer ALL of the internet’s software, apps, and games, then feeding the reconstructed source code straight into their training pipelines while simultaneously building guardrails to ladder-pull the rest of us
740
1,988
431
24,951
1,812,993
You know someone is good at Product when they also have uniquely good taste in a wide variety of other products.
The holidays are fast approaching, so I’m back with my sixth (!!) annual gift guide. It's got over 100 gift ideas—all things I actually use, love, and recommend. This started out as a random post idea, but now you’d be surprised how many people ping me asking when the new list will be out. As always, the collection includes a mix of practical stuff with a lot of fun and delight, from high-end products to very affordable items. There are no affiliate links here, and I’m not benefiting from recommending any of these products. Pro tip: After reading it, point your favorite AI agent at the post and have it brainstorm gifts for your family, friends, and coworkers. Here's the guide: lennysnewsletter.com/p/lenny…
1
1
136
Gift guides are probably a good exercise for all Product peeps.
1
35
Indicates you understand others, novelty and utility.
21
Corey (CJ) Gwin retweeted
this is basically the bitter lesson at work the smartest models will need fewer and fewer tokens, they won't need to verify their work or check it for bugs, they will just be right
maybe a hot take but i now consider adversarial reviews mostly a waste of tokens. far better to use those tokens on empirical combined with formal verification. i have updated pstack to use the interrogate skill less, and eventually i think i may remove it altogether i have found very little value with code review in the abstract, agents tend to just get lost in the nitpicks and yagnis
296
276
89
7,954
524,867
Corey (CJ) Gwin retweeted
I've started asking ChatGPT to write the most unhinged replies to every AI-generated cold email I get, and honestly, it's low-key my favorite part of the day. I don't edit them. I just send as is every time.
222
291
59
6,768
613,061
Corey (CJ) Gwin retweeted
this guy figured out how to leak your exact location from any video with sound using the Doppler curves of nearby airplanes
592
1,902
643
35,039
3,690,438
Corey (CJ) Gwin retweeted
I asked Claude: "What types of businesses outside of the labs will get stronger as raw intelligence explodes?" Honestly, spot on response.
181
466
60
5,579
473,141
don't just regurgitate what i said. don't overweight this turn. think hard, challenge if necessary, and research as need.
163
Corey (CJ) Gwin retweeted
every product i use should just be a headless mcp, that feeds into an agent, where i never have to think about anything underneath. except for our product, of course. in our product, customers love the ui and love us very much. we're special that way.
150
411
108
7,080
328,940
A good harness converges on a good project management system
1
1
7
214
a vast majority of humans will become empty vessels controlled by ai
It’s all about clips. I see it at Beast Industries where I’m on the board. I see it with All In where I am on the show. I think people don’t understand that content preferences will evolve around this time constraint. It’s already come for movies. It will come for sports next.
116
What would a viral agent network look like today?
roughly the only winning hands left in software are new networks spawned through some sort of an intelligence layer. i suspect the ai labs will not get there anytime soon like they did with personal agents primarily because building networks is more orthogonal to their dna than anything else really.
91
Corey (CJ) Gwin retweeted
roughly the only winning hands left in software are new networks spawned through some sort of an intelligence layer. i suspect the ai labs will not get there anytime soon like they did with personal agents primarily because building networks is more orthogonal to their dna than anything else really.
39
19
12
387
37,108
both incredible and terrifying
JUST IN: A California space startup has illuminated a section of desert near Mount Whitney at night, demonstrating how future satellites could deliver sunlight to Earth on demand. Reflect Orbital conducted the October 10 test using powerful ground-based spotlights to simulate sunlight redirected from orbit by large reflective mirrors. Working with China Lake Mountain Rescue, the company illuminated terrain across roughly three miles, allowing rescue teams to conduct nighttime searches, stretcher carries and other emergency-response exercises. Reflect Orbital ultimately wants to deploy satellites capable of directing sunlight toward specific areas on Earth after sunset, potentially supporting emergency operations and extending electricity generation at solar farms. The company received FCC authorization in July for its first experimental satellite, Eärendil-1, which will use an 18-meter reflector to test the concept from orbit. The October demonstration used ground-based lighting. No orbital mirror has yet demonstrated the proposed capability.
76