Hanna Huffman

👩🏻‍💻 Series C+ Growth Leader 🃏 Wannabe MTG Tournament Grinder 🤖 Building the playbook for Agentic Growth teams

[Hanna Huffman @bananahuff]

functional marketing roles are collapsing. emily kramer, jonathan martinez, and alex lieberman are all building around the same thesis from different angles.

i've been doing this for two years. today i published what's actually inside the system.

Link to Install In the thread.

[Cody Schneider @codyschneider]

you're competing against marketers who don't know what a MCP is

[Hanna Huffman @bananahuff]

@Austen I can’t ever get this to work. I have too many email accounts and the auth tokens need to be re-authenticated to frequently. What ended up working for you?

[Hanna Huffman @bananahuff]

@clairevo Feels great to see someone else say this. If you require cross country travel for every meeting, commit to in person.

[Hanna Huffman @bananahuff]

@aakashgupta Nothing worse than the “it worked this way 10 years ago so we keep doing it that way today”, especially in the post-AI era. 💀

[Hanna Huffman @bananahuff]

@rileybrown Not very different from Claude Chat and Code being in the same app window. Pretty easy / convenient IMO and already has user familiarity as precedent t

[Hanna Huffman @bananahuff]

@Gaus450 Browser control

[Hanna Huffman @bananahuff]

@Codie_Sanchez I worked at a company both pre- and post-HRBPs and can say that the bloated “people” functions played a major role in ruining that business.

[Hanna Huffman @bananahuff]

a client spent a year "doing AI."

automated a dozen things. pipeline is flat.

they're not behind on tooling. they're in the automation trap: pointing AI at hours saved instead of impact created.

new piece on the 80/20 loop, and why loops beat workflows every time 👇

[Hanna Huffman @bananahuff]

@petergyang Codex > Claude

[Hanna Huffman @bananahuff]

"be more creative with AI" is useless without examples.

so i made the examples: 15 creative growth plays that move pipeline, with the operator + the mechanism for each.

ad engines, SEO 2.0, trained writing models, 10 landing pages instead of 1.

reply LOOP and i'll send the lis

[Hanna Huffman @bananahuff]

hot take: paying frontier prices to do junior work is the most expensive habit in AI.

route each chunk to the cheapest model that can do it:
• RouteLLM: ~75% cheaper, ~95% of the quality
• tiered routing: ~85% savings
• OpenRouter Fusion: top-tier output, ~half the cost

it's

[Hanna Huffman @bananahuff]

Nadella this month:

"You can offload a task, or even a job, but you can never offload your learning."

every automation you build and walk away from banks a little time and teaches you nothing.

the teams winning with AI aren't automating more. they're building a learning loop.

[Hanna Huffman @bananahuff]

@petergyang Extremely slept on function

[Hanna Huffman @bananahuff]

Sometimes I’m feeling so behind in the post-AI era and I start freaking out. Then I see a client take 4 weeks to install GTM and realize the majority are still completely asleep.

[Hanna Huffman @bananahuff]

@garrytan And people seem to think that the answer is proctored, times Wonderlic testing. SMH

[Hanna Huffman @bananahuff]

@OpenRouter @grok ELI5

[Hanna Huffman @bananahuff]

I’ve worked for many startups where the founder has never even met most of the people that are helping them build their business — even at very senior levels that have been there for four more years. Sad to say, this is really special and probably explains a lot about why Tesla is so successful.

[Teddy - PolyBackTest.com @Bitcoin_Teddy]

Elon Musk: There are no lords and peasants at Tesla. Everyone eats at the same table.

“I actually know the people on the line, because I worked on the line, I walked the line, I slept in the factory, and I worked beside them. So, I'm no stranger to them.

There are many people at Tesla who have gone from working on the line to being in senior management. There are no lords and peasants. Everyone eats at the same table. Everyone parks in the same parking lot.

At GM, there's a special elevator only for senior executives. We have no such thing at Tesla.

We give everyone stock options. Many people who are just working the line, who didn't even know what stocks were, we've made them millionaires.

And I just want to say that I'm incredibly appreciative of those who build the cars, and they know it.”

New York Times DealBook Summit, 2023

[GREG ISENBERG @gregisenberg]

The takeaway from Fable 5 being BANNED by the government: GET GOOD AT LOCAL MODELS SO YOU HAVE 100% CONTROL.

My entire weekend was going to be building my craziest ideas with Fable 5. That's now cancelled.

So instead of building with Fable this weekend, I've decided I'll go deep on local models:

1. Start with the runtime. Download Ollama or LM Studio first. This is the thing that actually runs models on your machine.

2. Match the model to your hardware. A model's size is measured in billions of parameters (7B, 32B, 70B). Bigger is smarter but needs more memory. Rule of thumb: a 7B model runs on almost any laptop, a 32B needs a good Mac with 32GB+ RAM, a 70B needs serious hardware like a DGX Spark or a maxed-out Mac Studio.

3. Know which model for which job. Qwen 3 is the best all-around choice for most tasks. DeepSeek for reasoning and coding. Gemma 4 when you need something tiny that runs on a phone. Llama when you want the biggest community and the most fine-tunes.

4. Quantization. You can shrink a model to run on weaker hardware with barely any quality loss. Look for versions labeled Q4 or Q5. This is how a model that "needs" a server runs on your laptop. Learning this one concept changes everything.

5. Connect it to your agent. Point Hermes or your agent stack at a local model.

6. Context window is your real constraint locally. Cloud models give you huge context for free. Local models make you pay for it in memory. A bigger context window eats RAM fast. Keep your sessions tight and your prompts lean or your machine chokes.

7. Learn to give local models tools. A smaller local model with web search, file access, and code execution beats a giant model with none. The capability gap closes fast when you wire up the right tools. The model is the engine but the tools are the wheels.

8. Fine-tuning is more accessible than you think. You don't need this on day one, but know it exists. You can take an open model and train it on your own data so it gets good at your specific domain.

I'll probably do a breakdown at some point on this @startupideaspod if people are into it.

The lesson from this ban is basically don't build your entire workflow on something that can disappear with a single letter. Own part of your stack. Local models are insurance.

It reminds me when people realized they don't own social media accounts. And then you saw people build email lists etc.

I remember running a startup and my biggest traffic source was organic FB. All of a sudden, algo changed, and I lost 99% of my traffic.

Same sorta moment (but bigger) for AI.

This is a wake up call.

[Hanna Huffman @bananahuff]

@businessbarista @codyschneider @PJaccetturo @searchbrat @ericosiu Some funny choices here

[Hanna Huffman @bananahuff]

Truth hurts

[Paul Graham @paulg]

You are far more dangerous to your startup than competitors are. A hundred times more startups die from poor execution by their founders than are killed by competitors.