Json

#Author, #Photographer, and avid #coffee drinker #fractionalcto

[Json @notjasonmorris]

And it's published! We hope this guides anyone interested in how the command line helps with day-to-day data engineering/science tasks. Also, thanks to @PacktPub for the support!

@PacktAuthors

https://t.co/ECo8db4e7N

#datascience #commandline #data

[Json @notjasonmorris]

Prediction: in 24 months the default for new AI features is an open weight model you host. Frontier APIs are the escalation path.

59% of Chinese 20B+ open releases ship Apache 2.0. GGUF repos up 464%.

Build the seam now.

Full read on our LinkedIn.

#OpenWeights #SelfHostedAI

[Json @notjasonmorris]

Half the founders asking about a fractional CTO should not hire one yet.

New survey: median time to positive ROI is 90 days. Name what changes in 90 days. If you cannot, it is not a leadership gap. It is an unscoped problem.

More on our LinkedIn.

#FractionalCTO #StartupCTO

[Json @notjasonmorris]

Meta explored cutting some teams 60% to go AI native, then pulled back.

Internal data: infra changes up 220%, shipped features up 36%, incidents up 40%.

More motion, not more product. Pick the metric before the tooling.

Full read on our LinkedIn.

#AIAdoption #ProductVelocity

[Json @notjasonmorris]

A founder came to us shipping faster than ever, trusted less than ever by customers.

The AI tools were not the problem. Adopting them without moving the review bar was.

Speed without a review bar is not velocity. It is debt.

Full read on our LinkedIn.

#TechDebt #StartupCTO

[Json @notjasonmorris]

Before renewing frontier model spend, test one workload: log 500 real inputs from your highest volume AI call, fine tune a small open model, shadow it a week.

A 350M model, one epoch, beat chain of thought baselines on ToolBench.

Full read on our LinkedIn.

#SmallModels #MLOps

[Json @notjasonmorris]

Myth: real AI has to live in a datacenter.

Mobile NPUs now hit 35 to 60 TOPS, 2017 datacenter GPU territory. Local isn't weaker, it's bandwidth bound. That gets engineered away.

Audit the workload before you rent a GPU.

Full read on our LinkedIn. #OnDeviceAI #EdgeAI

[Json @notjasonmorris]

Open weight models are now months behind the frontier, not years.

But weights ship without guardrails. Download one and the safety work becomes your line item, not your vendor's.

Capability got cheap. Ownership did not.

Full read on our LinkedIn.

#OpenSourceAI #AIStrategy

[Json @notjasonmorris]

Startup hiring is only 4% below 2019. Big tech is down 25%. Small teams are winning.

But new grad hiring at startups fell 76%. The bench that grew your next tech lead is gone.

Leverage is judgment now, not headcount.

Full read on our LinkedIn.

#Startups #FractionalCTO

[Json @notjasonmorris]

On-device AI's real bottleneck isn't compute. It's memory.

Mobile NPUs now hit 35-60 TOPS. But phone memory bandwidth is 50-90 GB/s vs 2-3 TB/s in the datacenter.

Model size is a product constraint, not a research preference.

Full read on our LinkedIn.

#OnDeviceAI #EdgeAI

[Json @notjasonmorris]

Most seed startups do not need a full time CTO. They need CTO level decisions made right, early.

About 40% of C suite hires fail inside 18 months. At seed that is not a salary loss, it is a year of architecture to unwind.

Full read on our LinkedIn.

#FractionalCTO #Startups

[Json @notjasonmorris]

Founders reach for the biggest model. Then the invoice arrives.

Most agent calls are narrow: classify, extract, route. A small tuned model does that at a fraction of the cost and latency.

Most traffic never needed a frontier model.

Full read on our LinkedIn.

#SLM #AgenticAI

[Json @notjasonmorris]

Local AI's bottleneck was never compute. It's memory bandwidth.

4-bit quantization cut memory traffic 4x. Speculative decoding added 2-3x.

Summarization and extraction now run on hardware users already own. Datacenters are the fallback.

Full read on our LinkedIn. #LocalAI

[Json @notjasonmorris]

Open weight models now run 29% of AI Gateway tokens on 4% of the spend.

The best sit 3 to 6 months behind the frontier, a gap held for 18 months.

Most of your stack does not need frontier pricing. Build a routing layer.

Full read on our LinkedIn.

#OpenSourceAI #AIStrategy

[Json @notjasonmorris]

Junior eng hiring at early stage startups: down ~76% since 2019.

The cheap capacity that used to absorb bad architecture decisions is gone. What you build in year one, a tiny senior team carries for years.

Build for that. Full read on our LinkedIn.

#Startups #FractionalCTO

[Json @notjasonmorris]

Nearly half of phones shipping in 2026 can run AI on-device. Counterpoint puts GenAI-capable smartphones at 45% of global shipments this year.

Every inference you move local is one you stop paying per token for.

Full read on our LinkedIn.

#OnDeviceAI #EdgeAI

[Json @notjasonmorris]

Series A math nobody runs: a CTO search takes 6 to 9 months, then 3 to 6 more to ramp. A year of runway gone before the first architecture call.

Most founders do not need a CTO yet. They need senior judgment, faster.

Full read on our LinkedIn.

#FractionalCTO #Startups

[Json @notjasonmorris]

Most product AI work is not reasoning. It is classification, extraction, routing.

Task-specific small models can cut inference cost up to 90% on that work. Gartner expects 3x more SLM use than LLM use by 2027.

Full read on our LinkedIn.

#SmallLanguageModels #AI

[Json @notjasonmorris]

The biggest AI infra story of 2026 is not another gigawatt datacenter. It is how much inference no longer needs one.

Cost, latency and data sovereignty are pulling workloads local. The bottleneck now is memory, not models.

Full read on our LinkedIn.

#LocalAI #EdgeAI

[Json @notjasonmorris]

Apple just put a 20B parameter model on the iPhone. The trick: only 1 to 4B parameters activate per request. Sparse beats big. The on-device AI floor just moved, and so did your infra math. Full read on our LinkedIn. #OnDeviceAI #EdgeAI #Startups

[Json @notjasonmorris]

The average CTO search: 6 to 9 months. About 40% of C-suite hires fail within 18 months. Fractional CTO adoption has tripled since 2021. Founders need senior judgment now, not in month nine. Full breakdown on our LinkedIn. #FractionalCTO #Startups #TechLeadership