About
Passionate builder of great teams that reach high to achieve BHAGs.
Articles by Chip
-
The Election, Millennials and Trust – and What it Means for Retailers
The Election, Millennials and Trust – and What it Means for Retailers
The 2016 election is captivating millions of Americans, and passion amongst the electorate is arguably at an all time…
9 Comments
Activity
13K followers
-
Chip Overstreet reposted thisChip Overstreet reposted thisTokenmaxxing is a leading indicator with an expiration date. Most teams we talk to are already past it, they just haven't changed the metric yet. Early on counting tokens is fine. When you're trying to get a team to use coding agents token volume tells you something real. The engineer burning 50M tokens a week has changed how they work. The one at 500k hasn't. At that stage, count tokens and don't overthink it. The trouble starts once adoption sticks. When everyone is using the tools daily, raw counts stop measuring anything meaningful. A billion tokens on a small model and a billion on a frontier model differ in cost by more than an order of magnitude, so the comparison isn't real. At some point you're just measuring who left an agent running overnight. There's a better question for teams past that phase, especially on Max-style plans with fixed price and usage limits. You've already paid for the capacity, and whatever you don't use is just gone. So the question isnt "how many tokens did we burn" it should be "how much of value did we get out of what we already paid for?" We've been calling that value maxxing. Price each person's usage at API rates and compare it to what their plan costs. Carter Bastian ran this on himself recently and got 8.5x: $200 Max plan against ~$1,700 of tokens at API rates. My last 30 days was 9.8x. We also found a Codex seat on our own team at 0x, $200 a month for the plan, $0 of usage (oops). Raising your multiple is mostly about respecting the windows. Headroom expires, a five-hour window that resets with 60% unused is value you paid for and didn't collect. So kick off the heavy agentic work when a window is fresh, not right before it resets. When a window is about to expire with headroom left, spend it on work that's cheap to queue and easy to review later like test coverage, docs, discovery and spec work for the next project. And on a fixed-price plan there's no reason to economize on model choice. The strongest model costs the same as the cheapest one. Is value maxxing gameable? Of course, you can still burn tokens for the sake of burning them. But the waste is capped at money you already spent, and pricing at API rates means comparisons between people actually hold up. What we really like is that it points in both directions: a high multiple says the plan is a bargain worth protecting, and a sub-1x seat says downgrade it, or go find out why that person isn't getting value. A token count never tells you to spend less. We built this into Track. Each row on the leaderboard shows what a member paid, what their usage would have cost at API rates, and the difference. It also shows how much headroom is left in the current window, so prepaid capacity doesn't quietly expire. Anyone who doesn't want to be on the board can go anonymous. Setup takes a couple of minutes: brew install codelexica/tap/cdlx then cdlx init. Free, no credit card. Link in comments. #AI #EngineeringLeadership #DevTools #LLM #AICost
-
Chip Overstreet reposted thisChip Overstreet reposted this𝗟𝗮𝘀𝘁 𝗺𝗼𝗻𝘁𝗵, 𝗺𝘆 𝗔𝗜 𝘂𝘀𝗮𝗴𝗲 𝘄𝗮𝘀 𝘀𝘂𝗯𝘀𝗶𝗱𝗶𝘇𝗲𝗱 𝗯𝘆 𝟴.𝟱𝘅. I paid $200 for my Max plan. At API rates, those 2.1 billion tokens would have cost $1,711.89. Here is how I ran the numbers, and why your team should too. At Code Lexica, we think about token usage constantly. We are an AI-first startup: agentic tools run most of our operations, and our entire dev process is built around AI code generation. So we asked a simple question. If we dropped the subscription and paid pure usage-based pricing, what would our engineering actually cost? For one month, on one engineer, the answer was $1,711.89 against the $200 I paid. An 8.5x subsidy. Why it matters: tokens are artificially cheap right now. Pro and Max plans stack even deeper subsidies on top of already-low prices. The big model providers are running at a loss to win market share and get teams hooked. That will not last. Recent changes to plan structure and pricing suggest the subsidies are already drying up. Even if your usage is modest, it is worth asking: if the subscriptions disappeared next month, what would it cost to keep your team running? We solved this for ourselves, so we shipped it for everyone. This week we released Track, a free-forever tool that measures your team's real token usage and True Cost. Five minutes to set up: `brew install codelexica/tap/cdlx` then run `cdlx init` Point it at the directories you want to watch, and it captures usage from Claude Code, Cursor, and more. Every week you get a digest: the headline metrics, what changed, and why. For teams, break it down by member, project, model, or tool. Find out what your AI actually costs before the market decides to show you. Track is free forever, no credit card: https://lnkd.in/guJmuXVy #AI #EngineeringLeadership #DevTools #LLM #AICost
-
Chip Overstreet reposted thisChip Overstreet reposted thisI just released the DD Photos App - dougdonohoe/ddphotos-app at GitHub (https://lnkd.in/giQvy4Xa) - an open-source desktop frontend to my open-source 'ddphotos' static site generator. Use it to easily create your own super-fast photo galleries - with minimal technical knowledge (publish for free via Cloudflare pages). I use it for my own personal albums at photos.donohoe.info (https://lnkd.in/gVzyWguK).
-
Chip Overstreet reposted thisChip Overstreet reposted thisIf your team runs 'claude -p' on a subsidized Claude Max plan, your effective costs are going up 25x. Welcome to the end of the agentic free lunch. - Anthropic (June 15): Programmatic Claude usage: Agent SDK, claude -p, GitHub Actions, third-party tools moves off subscription pools onto a separate monthly credit billed at full API rates. Pro gets $20, Max 5x gets $100, Max 20x gets $200. Non-rolling. For anyone heavily running automations on a Max plan, it's an effective 25x cut. - GitHub Copilot (June 1): All plans move to usage-based billing. Same sticker price, but agent mode now eats from a metered pool. You'll pay the same but get a lot less In our customer discovery, most mid-sized engineering teams are running agents on subsidized subscriptions, not API tokens. A lot of that runs through claude -p and similar headless workflows. The economics of agent-driven dev have been quietly propped up by the labs. That prop is being pulled. Why now: Agents burn 10–1000x more tokens than chat. A single SWE-Bench Pro task on Opus 4.7 burns $10–$15 in API tokens for one PR in 15 minutes. Stanford research found ~70% of agent tokens are waste, re-reading files, exploring dead ends, snowballing context. Subsidies were a land-grab, the bill is now arriving. This isn't price gouging, it's the real cost of agentic work surfacing. 6 ways to cut your headless token bill starting today: - Trim your CLAUDE.md ruthlessly. It loads on every single claude -p invocation. A bloated file gets paid for on every run. Move specialized guidance into on-demand skills. - Pre-filter before the agent sees data. Don't make Claude read a 10,000-line log to find one ERROR. Grep, filter, and shape inputs in the wrapping shell -- pipe in the 12 lines that matter. - Ground the agent in pre-indexed context. Every claude -p call starts blind. No memory of yesterday's run, no map of your repo. Replace blind exploration with persistent, structured context, the cheapest token is the one the agent never spends rediscovering your codebase. - Route models per invocation. You can't /model mid-run in headless. Pick the right model up front: Haiku for triage, Sonnet for most execution, Opus only when it's earned. -Turn off extended thinking when you don't need it. Thinking is on by default and bills as output tokens. Most automation tasks like refactors, file ops, structured edits don't need it. Reserve thinking for hard reasoning, not boilerplate. - Cap and instrument every run. Use --max-turns, set hard spend caps, log token counts per invocation. There's no human at the keyboard catching a runaway loop. One bad cron can hit $1,500 in a day. The teams that thrive in this next phase won't be the ones with the biggest budgets. They'll be the ones who treat tokens like a resource. P.S. Shameless plug: Code Lexica (https://www.codelexica.com) tackles this directly with persistent, pre-indexed codebase context via MCP, cutting 15–30% off agent exploration token spend.
-
Chip Overstreet reposted thisChip Overstreet reposted thisFor the first time in the history of software, backlogs are actually getting drained. Devs can ship faster than PMs can spec what to build next, and the bottleneck has shifted. In our latest post, we break down what AI-first product workflows actually look like in 2026: • Why the grey area between intent and execution is where the next 10x in product velocity lives • What it takes to automate specs, tickets, and sprint planning without generating garbage • Why building this in-house is harder and riskier than most teams expect Read the full post: https://lnkd.in/gPHgDuDiAI Ate Engineering, Now It's Coming For Product ManagementAI Ate Engineering, Now It's Coming For Product Management
-
Chip Overstreet shared thisVery simply, this makes any AI coding agent and engineering team better, no matter the level of sophistication, without requiring changing out anything in their current dev environment. Engineering orgs in the early stages of AI adoption can get up and running quickly with Claude, Cursor, Copilot -- without fear of engineers going off the rails. Sophisticated teams already leveraging AI extensively get better context for every facet of the dev cycle. Give it a try!Chip Overstreet shared thisToday I’m excited to share what we’ve been building and working on at Code Lexica! Carter Bastian and I founded Code Lexica because we kept seeing the same problem: AI coding tools are incredible at generating code, but they have no idea how your codebase actually works. They read it, they index it, they generate code against it. But they don't get it, not the way your best senior engineer does. That engineer knows which services a change touches, what patterns to follow, and where all of the landmines are. That knowledge comes from years of building, breaking, and fixing the system. Every AI coding tool on the market operates without any of that. Every session, every prompt, every agent run starts from scratch. The tool ingests large swaths of code, makes its best attempt to discern how everything fits together, and generates output. For a small repo, fine. For the kind of multi-service platforms or legacy systems that most real engineering organizations maintain, it falls apart quickly. The uncomfortable truth is that as AI writes more of our code, fewer people deeply know the systems they're working on and the tools writing the code don't know them either. That's why we built Code Lexica, a codebase intelligence layer that gives every tool in your stack the deep context it needs to actually work. We've been working with early adopters to enable complex legacy migrations, automate AI engineering workflows, and help teams scope and navigate codebases with unprecedented speed and accuracy. We wrote up our thesis on where software engineering is headed and why codebase intelligence sits at the center of it. Read it here: https://lnkd.in/gB-DF4ST Would love to hear from anyone navigating this same shift. What's working, what's not, and where you're feeling the gaps.The Missing Intelligence Layer Between Your Codebase and AI ToolsThe Missing Intelligence Layer Between Your Codebase and AI Tools
-
Chip Overstreet shared thisThank you to the HSSA for a $750,000 grant to Blaze Barrier! In places like Spokane, wildfire smoke is cutting lives short - a report from the WA Dept of Ecology shows communities here face ~2.4 years lower life expectancy, driven largely by prolonged exposure to fine particulate pollution. If we can stop catastrophic wildfires at the source, we can materially reduce that burden. Appreciate HSSA for providing critical funding for all of these organizations that positively impact health outcomes at scale. Blaze BarrierChip Overstreet shared thisGreat News for Your Friday Afternoon- Spokane County's Health & Bioscience Industry Just Got Better- Congratulations to Our Newest Grantees! 🔗 Read All About It https://lnkd.in/g8U3sbak #HSSAimpact #AdvantageSpokane Blaze Barrier, Jacob Schuler, Jennifer Fanto Credential Network, Dylan Avatar Glyciome, LLC, Joanna Ellington, Gil Clifton Precision Quantomics, Chandima Bandaranayaka, Christy Watson Slate Flosser, Brynn MacLennan, Karen Plaister
-
Chip Overstreet shared thisProud to support Gonzaga University’s New Venture Lab and the incredible students who are working alongside local entrepreneurs to turn ideas into impact. I've been impressed by the energy and effort these students put in, but more so the output!
-
Chip Overstreet shared thisSo awesome! Did they just say "Spice Empire"?!! Well done Darby and team!Chip Overstreet shared thisAmazing to see the Spiceology story in Forbes - check out this great article and interview with our CEO Darby McLean: https://lnkd.in/gVnYu4xu
-
Chip Overstreet reacted on thisI’m especially proud to share this one. Eastern Washington is home for me, and it’s incredibly rewarding to see more than $35 million invested right here in the Spokane region through two projects that will have a lasting impact on our communities. Craft3 is investing $20.25 million to support the expansion of the Kalispel Tribe of Indians’ Camas Center Medical and Dental Clinic, and $15 million to support the new Salish School of Spokane campus. These are very different projects, but both represent the kind of community development I’m passionate about: expanding access and opportunity, strengthening communities, and investing in resources that will serve people for generations. I’m proud of the Craft3 team and the many partners who worked to bring these projects together—and personally excited to see this level of investment happening in the place I call home. This is what community development finance can look like when capital, partnership and community priorities come together. Proud of this work. Proud of our partners. And very proud to see it happening here in Eastern Washington. #EasternWashington #Spokane #CommunityDevelopment #CommunityFinance #Craft3 #NewMarketsTaxCreditsChip Overstreet reacted on thisMore than $35 million. Two Spokane projects. Two examples of what New Markets Tax Credits can make possible. Craft3 is investing: • $20.25 million to support expansion of the Kalispel Tribe of Indians’ Camas Center Medical and Dental Clinic • $15 million to support the new Salish School of Spokane campus Together, these projects will expand access to healthcare, education and cultural resources while supporting jobs and serving thousands of people. They also reflect how we believe community development finance should work: start by understanding what a community needs, then bring the right tools and partners together to help make it possible. Read more in the Spokane Journal of Business, which profiles the projects and Craft3’s work in the region. https://bit.ly/4yAYKap #NMTC #NewMarketsTaxCredit #CDFI #CommunityDevelopment
-
Chip Overstreet reacted on thisChip Overstreet reacted on thisIt’s taken me a few years back in software to figure out what I’m truly passionate about, and I finally have the answer: Lighting stuff on fire in front of prospects in a fancy restaurant. 🔥 Dinner-Led Growth is my brand-new product marketing consultancy focused on EDaaS: Executive Dinners as a Service. How am I different? If you’ve been to one of my dinners, you don’t need to ask. If you haven’t: this isn’t about basic logistics or ideation. I perform the dinner. High-value conversation mixed with theatrics, wit, wisdom, wordplay, and yes—occasionally fire. Now, for hire.
-
Chip Overstreet liked thisChip Overstreet liked thisTokenmaxxing is a leading indicator with an expiration date. Most teams we talk to are already past it, they just haven't changed the metric yet. Early on counting tokens is fine. When you're trying to get a team to use coding agents token volume tells you something real. The engineer burning 50M tokens a week has changed how they work. The one at 500k hasn't. At that stage, count tokens and don't overthink it. The trouble starts once adoption sticks. When everyone is using the tools daily, raw counts stop measuring anything meaningful. A billion tokens on a small model and a billion on a frontier model differ in cost by more than an order of magnitude, so the comparison isn't real. At some point you're just measuring who left an agent running overnight. There's a better question for teams past that phase, especially on Max-style plans with fixed price and usage limits. You've already paid for the capacity, and whatever you don't use is just gone. So the question isnt "how many tokens did we burn" it should be "how much of value did we get out of what we already paid for?" We've been calling that value maxxing. Price each person's usage at API rates and compare it to what their plan costs. Carter Bastian ran this on himself recently and got 8.5x: $200 Max plan against ~$1,700 of tokens at API rates. My last 30 days was 9.8x. We also found a Codex seat on our own team at 0x, $200 a month for the plan, $0 of usage (oops). Raising your multiple is mostly about respecting the windows. Headroom expires, a five-hour window that resets with 60% unused is value you paid for and didn't collect. So kick off the heavy agentic work when a window is fresh, not right before it resets. When a window is about to expire with headroom left, spend it on work that's cheap to queue and easy to review later like test coverage, docs, discovery and spec work for the next project. And on a fixed-price plan there's no reason to economize on model choice. The strongest model costs the same as the cheapest one. Is value maxxing gameable? Of course, you can still burn tokens for the sake of burning them. But the waste is capped at money you already spent, and pricing at API rates means comparisons between people actually hold up. What we really like is that it points in both directions: a high multiple says the plan is a bargain worth protecting, and a sub-1x seat says downgrade it, or go find out why that person isn't getting value. A token count never tells you to spend less. We built this into Track. Each row on the leaderboard shows what a member paid, what their usage would have cost at API rates, and the difference. It also shows how much headroom is left in the current window, so prepaid capacity doesn't quietly expire. Anyone who doesn't want to be on the board can go anonymous. Setup takes a couple of minutes: brew install codelexica/tap/cdlx then cdlx init. Free, no credit card. Link in comments. #AI #EngineeringLeadership #DevTools #LLM #AICost
-
Chip Overstreet reacted on thisOur team has spent the last few years quietly building something I believe this region has needed for a long time. A regional network and virtual incubator where AI practitioners, university researchers, industry leaders, and founders can connect and not just to talk about AI, but actually build solutions together. On July 23 we are taking the next step. Two panels. Honest conversations. Real partnerships. If you are responsible for developing your organizations AI strategy or solutions — your peers are here and we invite you to join them. 🎟️ https://lnkd.in/g_kyRcEi #InnovateINW #AI #Cybersecurity #InlandNorthwestChip Overstreet reacted on thisIf you are responsible for your organization's AI strategy — or you know someone who is — this is the event you do not want to miss! We are hosting Building the Region's AI Ecosystem on July 23 in Spokane Valley and we want the people in your organization who are thinking about AI, cybersecurity, and advanced computing to be there. This is not a general tech meetup. This is a structured conversation between the people who are building AI strategies, the researchers developing the solutions, and the startups putting it all into practice — right here in the Inland Northwest. Here is what you can expect: ▸ Panel 1 — Higher Education & Research Regional university researchers from Washington State University and University of Idaho and higher ed faculty from Eastern Washington University, Spokane Falls Community College, North Idaho College and others sharing active programs and capabilities in AI, cybersecurity, robotics, and emerging technologies. ▸ Panel 2 — Industry Challenges Regional industry leaders in energy, healthcare, manufacturing, aerospace, agriculture, and consumer industries sharing the specific challenges they are facing — and opening a direct dialogue with the researchers and innovators who can help solve them using AI, Cyber and Advanced Computing. ▸ Networking --> These are the conversations that turn into partnerships, internships, pilot projects, and solutions. Thursday, July 23 | 4:00 – 6:30 PM Burbity Workspace | 2818 N Sullivan Rd, Spokane Valley, WA 🎟️ Register here: https://lnkd.in/gYXANjeB Tag a colleague who needs to be in the room. 👇 #InnovateINW #AI #ArtificialIntelligence #Cybersecurity #InlandNorthwest #Spokane #Innovation #AIStrategy #DigitalTransformation #LaunchPadINW
-
Chip Overstreet reacted on thisChip Overstreet reacted on thisThe book I’ve been working on in the early mornings over the past two years has been accepted for publication! The book explores how several Columbine survivors experienced post-traumatic growth and developed resilience in the aftermath of tragedy. My hope is that their stories encourage anyone navigating a difficult season whether you’re leading others or facing a personal “lifequake.” I’m not sure of the release date yet.
-
Chip Overstreet reacted on thisChip Overstreet reacted on thisToday we announced Ultimus Fund Solutions as a client. Ultimus supports private equity firms, hedge funds, and other alternative asset managers, and are using our platform to centralize and automate payment processing and improve visibility into capital calls and other investment activities across their customer base. Bringing a fund administrator of this caliber into the Treasury4 community speaks to how well the platform handles multi-customer, high-volume environments. We are grateful to the Ultimus team for this client partnership. https://lnkd.in/gACedsud
Experience & Education
Recommendations received
6 people have recommended Chip
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content