Coffee With Developers Aug 17, 2026

Why Token Reduction Isn’t Cost Reduction - Dave Anderson & Sarel Weinberger, PhD.

Dave Anderson , Sarel Weinberger , Phd

Compressing LLM inputs doesn't save money. It forces models to work harder, driving up compute costs by 50%. Learn smarter resource allocation techniques that actually lower your bills.

Pause
Mute Enter Fullscreen
#1 about 6 min

The hidden costs of naive token compression

Compressing tokens upfront often forces AI models to do more reading and turns, driving up total computation costs.

#2 about 4 min

Establishing visibility for engineering team budget control

Engineering teams require full end-to-end telemetry to understand the financial impact and ROI of coding agents.

#3 about 6 min

Why most AI context remains unreachable for compression

System prompts, tool definitions, and internal reasoning occupy the vast majority of tokens that naive algorithms cannot optimize.

#4 about 3 min

Creating realistic benchmarks for standard coding agents

Standard benchmarks like SWE-bench misrepresent everyday coding tasks and skew the evaluation of model performance.

#5 about 8 min

How over-optimized context triggers excessive model computation

Removing seemingly irrelevant code causes models to lose confidence and execute expensive re-reads to verify context.

#6 about 8 min

Lowering costs by adjusting harness and effort levels

Utilizing cheaper execution loops and restricting loaded tool definitions drastically reduces unnecessary token generation.

#7 about 5 min

Combating deployment FOMO with practical cost tracking

Implementing a control plane helps teams avoid overspending on high-end models for trivial engineering tasks.

#8 about 6 min

Best practices for interacting with AI coding assistants

Avoid interrupting native harness loops and isolate tasks in fresh sessions to prevent exponential caching costs.

Matching moments

9:14 min

Assessing the token economy and costs of artificial intelligence

Chris Heilmann +2 · LIVE

2:09 min

Shifting focus from token maxing to AI cost management

Dona Sarkar +1 · Coffee With Developers

2:01 min

Transitioning toward AI-first coding and managing token costs

Clemens Wasner Clemens Wasner +4 · World Congress 2026 Europe

1:41 min

Calculating the hidden financial costs of AI coding assistants

Chris Heilmann +2 · LIVE

3:45 min

Optimizing token costs and intelligent chunking strategies

Luca Bianchi Luca Bianchi · World Congress 2026 Europe

3:21 min

Optimizing AI token consumption and running local language models

Chris Heilmann +2 · LIVE

Upcoming sessions on this topic

Open session

World Congress 2026 North America

Headroom: A Context Optimization Layer for LLM Applications

Tejas Chopra

Senior Software Engineer at Netflix

Tejas Chopra
Open session

World Congress 2026 North America

AI ROI: The Hard Unit Economics of AI-Native Engineering

Manu Gurudatha

Manu Gurudatha, VP of Engineering at PagerDuty

Manu Gurudatha
Open session

World Congress 2026 North America

public void saveMoney(AI): The Developer's Guide to Unit Economics

Hrushikesh Pokala

Senior Software Engineer Lead at Equifax

Hrushikesh Pokala
Open session

World Congress 2026 North America

Optimizing consumption and cost of AI APIs in the enterprise

Derric Gilling

Vice President and General Manager, API Platform

Derric Gilling
Open session

World Congress 2026 North America

Agents That Own Their Inference: Building Production AI Agents on Dedicated GPUs

Duan Lightfoot

Sr. AI Engineer, Akamai

Duan Lightfoot
Open session

World Congress 2026 North America

KV Cache Is Not About Speed: It's About Surviving Inference Costs

David vonThenen

AI/ML Leader | Keynote Speaker | OSS Engineer & Developer Advocate | Agentic AI, Deep Learning, Production AI | Python, Go, C++

David vonThenen