Owning Your Intelligence: Portable Memory, Model-Agnostic Harnesses
Today's top 12 insights for PM Builders, ranked by relevance from X, LinkedIn, and YouTube.
Owning Your Intelligence: Portable Memory, Model-Agnostic Harnesses
#1 đť•Ź
Harrison Chase recapped his “Owning Your Intelligence” talk, framing agents as model + harness + context and advocating portable memory plus model-agnostic, configurable harnesses. He also covered evaluations, observability, and a data flywheel that uses agent traces to drive improvement.
#2 đť•Ź
A post announced that Qwen3.8-27B is live on LM Studio and invited users to try it.
Also covered by: @Qwen
#3 in
Guillermo Rauch predicts that an unnamed command for configuring existing coding harnesses, including Claude Code and Codex, will become the default way to use coding AI at scale. He highlights uptime, model choice, lower costs, observability, and ZDR.
#4 in
🥞 Carl Vellotti shared a process for developing a Claude skill: complete one initial run with Claude and no skill, then have Claude codify the session as the specification. For text workflows, he suggests editing markdown output and storing lessons from those edits in a corrections folder the skill reads on later runs.
#5 in
Peter Yang announced an upcoming episode with Riley Brown (1.5M+ followers) about using Codex to run his content business. Brown demonstrates using Codex to find 100 top-performing thumbnails in his niche, add them to a Paper canvas, and combine them with photos of himself.
#6 ▶️
This new startup can query anywhere you've been...
Fireship
Flock Safety’s Falcon cameras use on-device machine-learning inference to create vehicle fingerprints and upload images plus structured metadata over LTE to Flock Cloud, where law-enforcement agencies can query nationwide vehicle-history data without warrants under the third-party doctrine.
- Founded in 2017 by Garrett Langley, Flock Safety is described as an $8.4 billion surveillance startup whose cameras scan billions of license plates per month across thousands of communities.
- The solar-powered Falcon camera uses an LTE modem, infrared night vision, a motion-triggered camera, and edge ML to classify vehicle make, model, color, dents, rims, roof racks, and stickers; it can identify vehicles even when license plates are missing or covered.
- DFlock, started by software engineer Will Freeman, is an open-source OpenStreetMap-based dataset that has located tens of thousands of Flock cameras; Flock’s lawyer sent Freeman a cease-and-desist letter, and he responded, “No.”
#7 đť•Ź
Thariq shared an artifact he made with Claude to help himself understand how watermarking without quality loss works, alongside a quoted @AnthropicAI post about an EU AI Act watermarking FAQ.
#8 đť•Ź
claire vo 🖤 commented that she maintains two main OpenClaws, one lifeguard OpenClaw, and one Codex tuned to connect via SSH and rescue both main OpenClaws—a setup she calls annoying to maintain and a labor of love.
#9 đť•Ź
Sebastian Raschka described watermarking as an additional selection mask at certain positions during top-p/k sampling. In his example, it randomly samples among 5 likely top-p choices using a seed dependent on the previous token.
#10 đť•Ź
Sebastian Raschka said changing the distribution is unnecessary: making the random seed depend on the previous token using a private key can enable specific, reproducible top-p or top-k sampling without altering the raw distribution.
#11 đť•Ź
Dario Amodei argued that AI structurally concentrates power and that regulation need not do the same, saying Anthropic backs policies that slow frontier labs while benefiting smaller competitors, including SB53’s $500M exemption and stricter testing for frontier models. He also supports the reported Trump administration approach to pre-deployment testing and Demis Hassabis’ idea for a FINRA-like entity, while cautioning that open-weight models alone are insufficient.
Also covered by: @Dario Amodei
#12 đť•Ź
Marko Slavnic created a fully animated short film using Runway, working alone for a few hours. The post says animation of this quality previously required a studio, a budget, and months.