Start here.

Selected guides for building a more capable business.

Mistral Large 4 and the business case for AI you control
Strategy

Mistral Large 4 and the business case for AI you control

Mistral’s Le Chonk brings a new candidate for AI you can run yourself. Here is what its planned open weights could change—and what a trillion-parameter model demands.

Read article
Where your AI runs should be your decision
AI ownership

Where your AI runs should be your decision

Claude Cowork’s Pro and Max cloud move highlights three choices: where workflows run, where models run, and where calls originate. With Looski, your team defines the deployment.

Read article
Should I sell my company's data to AI companies?
Opinion

Should I sell my company's data to AI companies?

AI labs and their data brokers are offering six figures for a company's Slack threads, email, tickets and code. The buyers have said what they want it for: training agents to do the work those records describe. For an operating business, a one-time check is a poor price for that.

Read article
Zapier vs Make vs Build Your Own
Compare: Automation

Zapier vs Make vs Build Your Own

Zapier vs Make vs Build Your Own: compare the fit, a focused custom workflow automation, ownership benefits, operating costs, and a practical pilot.

Read article
Quantization: memory savings, rounding error, and evaluation.
Inference explained

Quantization: memory savings, rounding error, and evaluation.

Work through an affine quantizer, include scale metadata in the memory budget, and evaluate the resulting model on the tasks you actually run.

Read article
Build an AI workflow your team can evaluate and maintain.
Field notes

Build an AI workflow your team can evaluate and maintain.

Define an output contract, build a held-out evaluation set, measure correction time, and release workflow changes with a rollback path.

Read article
Designing a testable data boundary for private AI.
Privacy

Designing a testable data boundary for private AI.

Map documents, embeddings, caches, logs, and integrations. Then test egress, access isolation, deletion, and recovery with synthetic records.

Read article
Unified memory: budget weights, context, and concurrency.
Apple silicon

Unified memory: budget weights, context, and concurrency.

Estimate model storage, KV cache, runtime overhead, and headroom separately. Then benchmark the workload at the concurrency you plan to serve.

Read article