Technical AI newsletter for engineers building on AI systems, covering evals, inference, and agentic architectures.
In plain terms: This technical newsletter and podcast covers how leading teams build artificial intelligence models, agents, and infrastructure. You can use it to learn how modern AI systems are built and keep your engineering knowledge up to date.
Why I recommend it: If you want an AI engineering job, this is the vocabulary you need.
Read / subscribe freeThis is the only entry from Latent Space by Swyx & Alessio in the hub so far.
Other resources that share this publication's topics.
Medium CEO Tony Stubblebine's announcement of TK, a private agentic notebook app from Medium for drafting writing and developing ideas, describing the company's position that writing still matters as AI tools spread.
A DataCamp guide comparing GitHub Copilot training options for engineering teams, with published list prices verified October 30, 2026. DataCamp sells some of the training it lists.
A KDnuggets guide by Vinod Chugani that goes round by round through current data science hiring processes, including AI-assisted coding rounds and generative AI evaluation questions, and lists ways candidates can prepare.
A Built In survey of job seekers on how AI tools shape which employers they find and consider. Among respondents who use AI for employer research, 66% say they applied to a company they had not considered and 53% say they decided not to apply somewhere because of what AI showed them. Built In sells employer branding services.
Artificial Analysis reports Muse Ember benchmark scores and task-cost estimates, including an Intelligence Index score of 36 and $0.14 per task in its evaluation. These are evaluator-specific results, not a verified Meta launch announcement or evidence of free model access.
Epoch AI explains an Inference Performance Index for estimating chip capacity across agentic workloads and output-speed requirements. Its modeled comparisons, expressed in GB300 equivalents, depend on workload and hardware assumptions and are not universal real-world performance guarantees.
