WeAreDevelopers Logo WeAreDevelopers
Sign in
For You LIVE
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

57 sec

Introduction to token compression and Headroom Labs

Devanshi Vyas Devanshi Vyas · World Congress 2026 North America

9:14 min

Assessing the token economy and costs of artificial intelligence

Chris Heilmann 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

2:10 min

Real-world token savings across different AI agent workloads

Devanshi Vyas Devanshi Vyas · World Congress 2026 North America

1:41 min

Calculating the hidden financial costs of AI coding assistants

Chris Heilmann Chris Heilmann +2 · LIVE

More from Coffee With Developers

Using Agents to Curate the Content You See Online - Dan Gish thumbnail
12:47
about 2 months ago

Using Agents to Curate the Content You See Online - Dan Gish

Dan Gish

AI Vector Search at Scale - Ewa Szyszka - Ewa Szyszka thumbnail
11:14
about 2 months ago

AI Vector Search at Scale - Ewa Szyszka - Ewa Szyszka

Ewa Szyszka

What is Cross-Origin Storage - Thomas Steiner thumbnail
34:51
about 2 months ago

What is Cross-Origin Storage - Thomas Steiner

Thomas Steiner

Women in AI - Nalini Garg thumbnail
42:19
about 2 months ago

Women in AI - Nalini Garg

Nalini Garg

Agents and People Dynamics in Shared Chats - Sam Liu thumbnail
37:41
about 2 months ago

Agents and People Dynamics in Shared Chats - Sam Liu

Sam Liu

Validating Real-World Usefulness of AI Models - Julia Kasper thumbnail
22:00
about 2 months ago

Validating Real-World Usefulness of AI Models - Julia Kasper

Julia Kasper

Upcoming sessions on this topic

Open session

AI Is the Easy Part: Why Faster Engineers Don't Make Faster Businesses

  • Bharat Sharma

    Booking

    Senior Engineering Manager

Open session

Teaching AI Coding Agents to Build It Right the First Time

  • Sajeetharan Sinnathurai

    Microsoft

    Principal Product Manager

Open session

Beyond File Dumps: Context Engineering for Coding Agents

  • Animesh Dutta

    Arm

    Senior Software Engineer

Open session

Prompt Debt: Why Your AI System Is Frozen in Time

  • Suriya Prakash B

    Red Hat

    Principal Software Engineer

Open session

Build a Zero-Cost AI Agent in Your Browser

  • Aishwarya Mathuria

    Adobe

    Computer Scientist

  • Nishant Thakur

    Adobe

    Senior Computer Scientist

Open session

Beware of Strangers Bearing Code: Open Source Trust in the Agent Era

  • Vikram Vaswani

    Consultant

Related videos

Headroom: A Context Optimization Layer for LLM Applications thumbnail
27:37
World Congress 2026 North America • 20 days ago

Headroom: A Context Optimization Layer for LLM Applications

Devanshi Vyas

From minutes to seconds: Lessons learned optimizing multi-turn agentic workflows thumbnail
42:44
Europe 2026 Virtual • 4 months ago

From minutes to seconds: Lessons learned optimizing multi-turn agentic workflows

Douglas Reiser

Cutting LLM Costs Without Cutting Quality: How to Beat Proprietary LLMs with Fine-Tuned Open Source thumbnail
29:50
World Congress 2026 Europe • 3 months ago

Cutting LLM Costs Without Cutting Quality: How to Beat Proprietary LLMs with Fine-Tuned Open Source

Viktoria Semaan

ZTA: Zero Token Architecture thumbnail
30:36
World Congress 2026 North America • 20 days ago

ZTA: Zero Token Architecture

Kelsey Hightower

Unpredictable costs of AI vs predictable cost of Humans thumbnail
29:52
World Congress 2026 Europe • 3 months ago

Unpredictable costs of AI vs predictable cost of Humans

Hung Lee, Leandro Gomes da Silva, Raffi Krikorian, Anna Ott

Measuring the Wrong Things Faster Than Ever thumbnail
30:47
World Congress 2026 Europe • 3 months ago

Measuring the Wrong Things Faster Than Ever

Laura Tacho

Related articles

From learning to earning

Jobs that call for the skills explored in this talk.