Michael Jeffords

Co-Founder of Conifer (YC26)

[Michael Jeffords @MichaelJeffords]

It seems like everyone I talk to recently has been wanting to run different models through the same TUI/CLI. Feels like confirmation of something I've believed for a while, we need a definitive, model-agnostic third-party interface

[Dima Vremenko @DVremenko]

Claude Code TUI + gpt-5.6-sol.

Why settle when you can have the best of both worlds?

[Michael Jeffords @MichaelJeffords]

@thesupermanmx At what decode speeds though…

[Charles @charles_v11]

Pretty frustrating using Claude this month. Models seem slower, constantly hitting guardrails, and burning tokens. Planning is still superior, but definitely stresses the need for multi model setups

[Michael Jeffords @MichaelJeffords]

Lots of nuance to this question, while these are all great I find myself increasingly pointing others to LFMs. On a M3 36G I get ~500 tok/s on LFM2 350M with surprisingly good outputs

[Jun Song @jun_song]

Best Local AI models by VRAM size (7/18) :

Phone (~4gb) : Bonsai-27b
Nano (~12gb) : Gemma4-12b
Micro (~36gb) : Qwen3.6-27b
Small (~128gb) : Deepseek V4 Flash ds4
Medium (~256gb) : GLM-5.2-2bit
Large (~512gb) : GLM-5.2-NVFP4

[Michael Jeffords @MichaelJeffords]

When talking about the future of running on device a crucial question that remains is whether distillation/quantization of a large model extracts more value than routing between a series of small and hyper specialized local models

[Michael Jeffords @MichaelJeffords]

@TheAhmadOsman Am a firm believer that the future of running on device is not a single model with fable/kimi k3 level intelligence but rather routing between hyperspecialized local models that additively combine to rival frontier capabilities

[Michael Jeffords @MichaelJeffords]

@ivanfioravanti @TheAhmadOsman I think a future in which this is achievable is routing between fragmented hyper specialized local models, not a single quantized model that possesses fable level intelligence

[Michael Jeffords @MichaelJeffords]

@jun_song @jun_song As a single model with mythos level capabilities or routing between a mixture of fragmented hyper specialized local models?

[Charles @charles_v11]

Kimi K3 Benchmarks - seriously where is for profit in 18 months

[Michael Jeffords @MichaelJeffords]

@Tech2Wild Am curious to know the logistics of this, are these taken on hyper specific datasets that preserve intelligence?

[Michael Jeffords @MichaelJeffords]

Super excited to announce our launch of Juniper! Download and start saving on token spend!

https://t.co/EyJgOtaPbZ

[Conifer (YC S26) @coniferbuild]

Stop paying cloud prices for every AI request. We route 80% of your queries to your own hardware at no cost, using the cloud only for your most complex tasks. Available now @ https://t.co/Xj83Zhxf6r.

[Michael Jeffords @MichaelJeffords]

@DataChaz The pace of open models has been unbelievable. Even thrashing at 1 tok/sec doesn’t take away from the fact that less than a year ago, ~30B was the ceiling for a machine this size. Now it’s a 744B MoE.

[Michael Jeffords @MichaelJeffords]

We usually define AI permissions as files and APIs, but compute is a permission too. A study of 20 open-source agents found resource-exhaustion bugs in 16 of them. An agent should have a budget for CPU, memory and time, not just a list of tools.
https://t.co/2i4pLvpS8X

[Michael Jeffords @MichaelJeffords]

One misconception I keep seeing is that more parameters automatically means better performance.
A great example:
Google showed that a smaller on-device model outperformed large multimodal models at
intent extraction once the task was broken into steps.(https://t.co/uxHSiURoOW)

[Michael Jeffords @MichaelJeffords]

Got an interesting DM on Linkedin about @coniferbuild thought I'd share.

Hi Michael, congrats on @ycombinator S26.
One tricky post-accelerator question: when there’s interest, pilots, or early users, how clear is it who the real paying customer is?
Curious how you’re thinking about that.
Best,
------

Hey ------! We think about this question in two different horizons.
Near term, our real paying customer is B2B, specifically large enterprises. They carry the highest inference spend and the hardest data-residency requirements, so they're the ones with a clear reason to keep paying for Conifer. When we launch we'll get plenty of individual downloads, but only a small slice of those users have enough monthly API spend for the savings to matter. Post-launch retention cohorts will make the split obvious, and that's exactly how we'll separate curious users from actual buyers.
Long term, we're betting on the shift from cloud inference to edge. Local models keep getting smaller and more specialized through distillation and quantization. A setup where a router coordinates a handful of small on-device models to match one large cloud model on most tasks looks increasingly plausible. Once that crossover happens, the quality gap no longer outweighs the privacy and cost advantages of running locally, and our primary customer becomes the everyday consumer.

[Charles @charles_v11]

Fresh essay on scaling laws.
Sara Hooker spent years inside DeepMind and Cohere watching scaling laws get treated as gospel. Then she wrote down the heresy:
Falcon 180B was state of the art in 2023. One year later it lost to a model 22x smaller.
The $700B question isn't whether scaling works. It's whether it's still the cheapest way to buy intelligence.
She was convinced enough to quit and build a company on the answer.

[Michael Jeffords @MichaelJeffords]

Excited to share that @coniferbuild is part of the @ycombinator S26 batch. Looking forward to getting to work with @gustaf!

Token costs are eye-watering. @charles_v11 and I burned through $13,000 in Claude credits in just five days, and that's a fraction of what every company transitioning to Al is facing.

But it's not just the cost. Al usage today is fragmented: five different models, three subscriptions, three API dashboards, and a drawer full of API keys. Every team is flipping between tabs, juggling providers, and paying full price for all of them.

Conifer replaces all of it with one interface. We're the inference gateway that routes every Al query to the cheapest model that can actually do the job. Built on top of our Typhoon engine, which allows local models to run faster and punch far above their weight, Conifer only reaches for the cloud on your hardest tasks. The result is a token bill >80% less.

We launch tomorrow, July 7th, on https://t.co/weRgjPtW0x

If you're part of a company spending thousands on Al every day, been holding off on the switch because of cost, or someone just trying to decrease their spend, we'd love to hear from you. Email [email protected] or reach out here on X.

[Michael Jeffords @MichaelJeffords]

Completely agree, per-token prices keep dropping, yet reasoning models and agentic loops burn 10–100x more tokens per task, so the real cost of deploying frontier intelligence at scale continues to rise. The models are already smart enough, what's scarce is running them affordably.

[Charles @charles_v11]

So many great points, but the error-correction piece is the one people will underrate. LeCun's (1−ε)ⁿ doom assumes per-step errors are iid and absorbing, which they're not. Trained agents learn a restoring force back so long horizon behavior looks like a mean reverting walk. 👏

[Loc @locbuilds]

I have to say, I agree. I've been reading a lot of research on LLM routing, and it has consistently shown that the majority of queries don't actually need a frontier model. Most of what teams send to Fable 5 would come back just as good from something far cheaper.

And even for the most difficult prompts, Anthropic literally just shipped Sonnet 5 last week. It's nearly on par with Opus 4.8 (beats it on some knowledge work benchmarks) at less than half the price. Defaulting everything to the "best" model in 2026 shouldn't be a requirement, and oftentimes (as you pointed it out), it can be costly.