Finding what a task used
Before — Guessing at cost from token counts and tool estimates
After — Reading the session report and provider dashboard — and knowing what each number really is
3-hour hands-on workshop
Get More from Your AI Coding Budget
A hands-on cost optimization workshop for AI coding agents.
Learn where coding-agent cost really comes from, then optimize model choice, context, sessions, and tool output without sacrificing verified results.
Thursday, November 12, 2026 · 17:00–20:00 Europe/Sofia (EET, UTC+2)Online
Milko Slavov & Yavor BoychevMeet the instructors — live Q&A throughoutWhat this actually is
A working session for practitioners, not an AI overview.
This workshop is for people who already use coding agents and want it to cost less. Most of your budget disappears before any useful change happens. Invisible context (instructions, tool definitions, history, file reads, command output) competes for the same budget. Running every step on the strongest model wastes money directly, since most steps don't need it. Running everything on the cheapest model wastes money indirectly, through retries and rework when it gets things wrong. The workshop covers the core principles of where the cost goes. Then we'll put them into practice: measure the baseline, make the optimizations (model choice, context, sessions, tool output), and review the outcomes together.
Theory
The core principles behind where the cost goes, and how each part of the pipeline burns budget.
Baseline
One task, real usage numbers, recorded before anything changes.
Optimize
Choose the right model for the task, compress command output, explore the codebase faster and cheaper, etc.
Verify
Run the same task again: cost should be lower, quality shouldn't slip.
What you'll be able to do
Get more verified work from the budget you already have.
- 01You'll be able to read what fills an agent's context — instructions, prompts, files, tool output, history — so you can see where the budget goes before changing anything.
- 02You'll be able to find what a task really used — real charge, tool estimate, monthly-plan usage, invoice line — so you can stop guessing from token counts.
- 03You'll be able to match the model to the step — strong for planning and hard problems, cheaper for routine implementation — so you can cut spend without paying it back in retries.
- 04You'll be able to decide whether to keep, restart, or compact a long session — so you can preserve the goal, constraints, decisions, and test results instead of losing the plot.
- 05You'll be able to strip avoidable context — noisy command output, broad codebase reads, whole documents — so you can verify the cheaper run is still correct.
- 06You'll be able to compare before and after on usage or cost, correctness, speed, and review effort — so you can take one small, safe experiment back to your real workflow.
After the workshop
What changes on the next day.
Choosing a model
Before — One model for everything, or the cheapest one by default
After — Knowing the tradeoffs of every model choice, cheap or expensive, and picking on purpose instead of by default
Running long sessions
Before — One endless chat that slowly loses the plot
After — A deliberate keep, restart, or compact decision that preserves goal, constraints, decisions, and test results
Feeding the agent context
Before — Broad reads, raw command output, whole documents
After — Targeted code exploration, cleaned-up output, and efficient document reads, with the cheaper way verified still correct
Workshop leads
Taught by people who pay these bills themselves.
Founder, Felan
Milko has spent 20+ years as an engineer, Staff Engineer, architect, and CTO — including more than a decade at VMware on cloud services, platform architecture, and Kubernetes-based infrastructure — and now builds autonomous SDLC agents at Felan, where he pays these same token bills. He built this workshop because every cost guide he found was vendor marketing or unmeasured tips; this one starts from a measured baseline and proves each optimization in a shared repo.
Co-founder, Felan
Yavor brings 20+ years as a hands-on engineer and engineering leader from VMware, Docker, Mirantis, Carbon Black, and OfficeRnD — cloud automation, SaaS platforms, developer tooling, and distributed teams. He teaches the session-management and model-selection material he wishes his own teams had: practical trade-offs between cost, speed, and quality that you can defend to your manager.
Who it's for
Already using coding agents. Ready to spend smarter.
Software engineers using coding agents
- Working with Claude Code, Codex, Cursor, Copilot, Felan, or another coding agent
- Want to reduce wasted usage without sacrificing correctness
Senior and staff engineers, tech leads
- Standardizing agent workflows for a team
- Need defensible trade-offs between cost, speed, and quality
Platform, DevEx, and AI-enablement engineers
- Responsible for tooling and usage across teams
- Need to see where the budget goes and what changes it
Hands-on engineering managers
- Need to understand cost, quality, and rollout trade-offs
- Want one experiment to take back instead of a generic policy
Who it isn't for
- Complete beginners to programming or Git — the exercises assume you're comfortable in a terminal.
- People looking for an AI strategy keynote — this is a hands-on working session, not a talk.
- Teams shopping for prompt libraries or model benchmarks — we optimize your workflow, not leaderboard scores.
Expected starting point: Participants should be comfortable with Git, a terminal, and at least one AI coding tool. No machine-learning experience or prior knowledge of token billing is needed.
Workshop materials
Take the workshop home with you.
Participant guide PDF
The reference for everything covered, yours to keep and share with your team:
- Token economics fundamentals from the workshop
- How to find cost and usage information for API and subscription plans
- The tools and approaches covered, what each changes, and links to tools and provider docs
Session recording
Every attendee receives the session recording afterward, along with everything else on this page.
Demo repository
The same repository used for every task during the workshop, so each before-and-after comparison stays reproducible after the event.
Workshop agenda
Three hours from guessing to measuring.
Short explanations, live before/after demos, guided exercises, and real measurement, with Q&A throughout.
- Presentation
What are we measuring?
- Pick one example coding task and agree what "done well" means
- Track correct result, time spent, retries, and usage or cost
- Know whether you're on pay-per-use (API) or a monthly plan, since that changes what the numbers mean
- Presentation
Token economics for coding agents
- What fills the context: instructions, prompts, files, tool output, history
- What input, output, and cached tokens mean
- Why fewer tokens does not always mean lower cost
- Demo & exercise
Find what one task used
- Run a task, then find its usage report in the coding tool and provider dashboard
- Tell apart a real charge, an estimate, and monthly-plan usage
- Note what you cannot see; short Q&A at the end
- Break
Break
- 10 minutes.
- Demo & exercise
Use the right model for each step
- Easy vs hard tasks and when to switch back to the strong model
- Plan with a strong model, implement with a cheaper one
- Demo: handing work from one model to another with Prewalk
- Demo & exercise
Long sessions: keep, restart, or compress?
- When to continue a session, start fresh, or compact (summarize) it
- What can break when you compact
- What to keep: goal, constraints, decisions, changed files, test results, next step
- Break
Break
- 10 minutes.
- Demo & exercise
Cut wasted context
- Clean up noisy terminal output
- Filter output with ctx-wire and RTK, and what Headroom does differently
- Find code without reading everything (Codebase Memory) and read a document (MarkItDown)
- Presentation
Did it get cheaper and still correct?
- Compare before and after: correct result, usage or cost, speed, failures, review effort
- Write down one small change to try at work, how to tell it worked, and when to revert it
- Final Q&A; questions welcome anytime during the workshop
Why this workshop is different
Every cost tip has a catch.
We try each one on a real task, so you can see when it saves money and when it ends up costing more.
Use a cheaper model
It costs less per step, but it gets hard tasks wrong more often, and you pay again in retries and review. We show which steps a cheaper model can handle and when to switch back.
Use fewer tokens
Not every token costs the same. Context the provider has already seen is cheaper, as long as its start stays the same, so a shorter prompt can cost more. We show where the bill really comes from.
Compact long sessions
A summary costs less than the full history, but it can drop the goal, the constraints, or decisions you already made. We show what to keep, and when starting fresh is better.
Cut noisy output
Less terminal output means fewer tokens, but if the error gets cut too, the agent guesses and tries again. We try output filters on the same task and check the result still works.
Tickets
Secure your spot
Thursday, November 12, 2026 · 17:00–20:00 Europe/Sofia · Online. Early bird pricing ends October 2 — or sooner if the tier sells out.
Thursday, November 12, 2026 · 17:00–20:00 Europe/Sofia (EET, UTC+2) · Online
Early Bird
Early bird pricing
- 3-hour live workshop
- Demo repository
- Participant guide PDF
- Session recording
- Live Q&A throughout
€145
Price includes VAT
Regular
Regular pricing
- 3-hour live workshop
- Demo repository
- Participant guide PDF
- Session recording
- Live Q&A throughout
€197
Price includes VAT
Late Bird
Last chance pricing
- 3-hour live workshop
- Demo repository
- Participant guide PDF
- Session recording
- Live Q&A throughout
€245
Price includes VAT
Secure card checkout via Stripe — team discounts apply automatically by quantity. All prices include VAT.
Team tickets
Save up to 30%
Discounts start at 3 engineers, applied per engineer at the current tier price.
Buy team tickets
Choose your team size — your discount applies automatically at checkout.
Ticket sales have ended.
Frequently asked questions
What people ask before the workshop.
Be comfortable with Git, a terminal, and at least one AI coding tool. You don't need machine-learning experience or prior knowledge of token billing — the workshop starts from a measured baseline and builds up from there.
The concepts apply to any coding agent; the workshop speaks to people using Claude Code, Codex, Cursor, Copilot, Felan, or similar tools. Exercises run in one shared demo repository so before-and-after comparisons line up for everyone.
You'll use your own coding agent during the exercises, with your own API key or subscription — any usage cost during the exercises is yours. A setup checklist will be sent before the workshop so the first exercise starts on time.
Git, a terminal, and your coding agent of choice. A setup checklist will be sent before the workshop so the first exercise starts on time.
The workshop uses one shared demo repository so every optimization is demonstrated and measured on the same codebase. The final session helps you map what you learned onto one real workflow of your own.
Yes. The workshop is live and interactive, with guided exercises and Q&A throughout, and every attendee receives the session recording afterward.
All sales are final — there are no cancellations, returns, or resales. If the event is cancelled or rescheduled to a time that doesn't work for you, you receive a full refund of the purchase price minus any banking and transfer fees.
Yes — teams of 3 or more save 8–30% depending on size, applied automatically by quantity at checkout on the current tier price. Private cohorts for a single team are also available.
Explain that it's a three-hour, hands-on workshop about understanding and optimizing AI coding-agent usage, and that you'll bring back the participant guide and demo repository to share the relevant practices with the team. Copy, adapt, and send one of the templates below.
Need manager approval? Copy, adapt, send.
Email template
Subject: Approval request: Get More from Your AI Coding Budget workshop
Hi [Manager], I'd like approval to attend Get More from Your AI Coding Budget on Thursday, November 12, 2026 (17:00–20:00 Europe/Sofia, online). It is a three-hour, hands-on workshop about understanding and optimizing AI coding-agent usage. This is relevant to [our team/project] because [briefly describe how you use coding agents or the usage/cost problem you want to address]. The workshop covers: - How agent context, input, output, and cached tokens affect usage. - How to find the cost or usage information available through API and subscription plans. - Model selection, session management, and tools for reducing unnecessary context. - Practical demonstrations using one demo repository. Ticket: [price] Registration: [link] I will use the participant guide PDF and demo repository to share the practices most relevant to our team. Can this be covered by our [learning and development/team training] budget? Thanks, [Name]
Slack or Teams template
Could I get approval to attend Get More from Your AI Coding Budget on Thursday, November 12, 2026 (17:00–20:00 Europe/Sofia, online)? It is a three-hour, hands-on workshop about understanding AI coding-agent usage and reducing unnecessary context through better model, session, and tool choices. The ticket is [price]: [link]. I will use the participant guide and demo repository to share the relevant takeaways with the team afterward.
For teams
Bring the savings back to your team.
The participant guide and demo repository are yours to keep and share. Teams of 3 or more save 8–30% depending on size — tell us your team size and we'll apply your discount. Or run the workshop privately for your team.