Skip to content

Standard Reader

mjsdotsocial

0Readers

5Posts

todayLast active

From tokenmaxxing to self-replication

We had tokenmaxxing, then the tokenpocalypse, and then the token famine. This drove price wars, interest and competition in open source/weight models, new model types (e.g, decision, CDM, etc.), and the pursuit of locally-runnable models. As innovation, including LLM gateways like OpenRouter, continues to drive capability, capacity, and cost efficiency for the hosted model ecosystem, the same is happening for local models, and not just for running open models (e.g., Qwen Coder, etc.), locally on...

Pay attention to the AI company shakeout

Pay attention: as consumer-play AI companies like Character.ai (who filed for bankruptcy after ~5 years) are winnowed out, watch for AI startups to pivot toward the enterprise, or abandon their own stacks in favor of companies chasing the enterprise like Anthropic or OpenAI, who have the focus, innovation, and scale, to succeed. It's a familiar story.

Another new model architecture just dropped

Another new model architecture just dropped. According to ScaleDown, Constrained Decoding Models (CDM) are task-specific, purpose-built language models, each trained to do one job: compression, summarization, extraction, and classification. Check out the post from founder Neal Patel: https://lnkd.in/gbDE-yWn

You better move quickly

If you want the agentic runtime your team is developing to stay at parity with those offered by AI companies, you'll need to move quickly. Always-on, user-dedicated autonomous agents (Spark, Muse, Dots, Grok Bot, etc.), dynamic work analysis with model autorouting (monomodal, multimodal, omnimodal; decisioning, etc.), cost management (frontier, flagship, mid-tier, lightweight, open weight/source, etc.), data (ontologies, semantics, taxonomies, etc.), DBs, (relational, vector, graph, KV, multimod...

Your agentic runtime is already antiquated

The enterprise agentic runtime your team is building was likely built on a model of how people work in the enterprise: handling individual tasks. That evolved to serial chained tasks, and then, maybe, to semi-autonomous workflows where the agent decides what to do based on the input and training of your SMEs. The AI companies have already evolved past task-based workflows. Whats next? Always-on agents, an enterprise evolution of OpenClaw. Backed by a persistent user-dedicated secure VM or worksp...

You've reached the end.