AI and the Billable Hour

Jonathan: Hello and welcome to Ditching Hourly. I’m Jonathan Stark. Today I’m joined by Shawn Yeager. Shawn, welcome to the show.

Shawn: Appreciate it. Thank you, Jonathan.

Jonathan: Before I have you introduce yourself, I want to let the listener know we’re going to be talking about AI and its effect on the billable hour today. So with that said, could you let folks know a little about who you are and what you do?

Shawn: You bet. The through line for my career has largely been emerging technology and getting it to market. Starting back on Microsoft’s first browser team, which dates me, and moving through the SaaS and cloud wave, mobile, and for the last five years, Bitcoin. So I’m on my fifth wave of technology in AI. My background is computer science, but I got pulled early into being a suit, as it were. It’s been sales, marketing, partnerships: how do you take a technology and translate it into something customers see value in and want to part with their capital for? That took me from Microsoft to Accenture to early-stage startups to my own, including my work in consulting. Now I run a firm called Upshift, and we’re focused specifically on helping professional services firms cross this new chasm, to understand what you sell after AI.

Jonathan: You sent over a relatively recent article this morning. Let’s tee it up there. The reason AI is bad for hourly billers is that they sell their time, and AI gives you less to sell. A productivity increase is actually bad for you. But the insight I really liked from the article is that it pulls apart two things that used to be wrapped together in the billable hour. So take it from there.

Shawn: That piece came in response to a talk I’ve been giving called “AI Killed the Billable Hour.” It’s a purposely agitating title, to get a conversation going with my primary audience: professional services firm owners and senior partners. Billing by the hour will probably outlive us all, but the billable hour as a unit of value has now been exposed. What was hiding in it is now on display as AI crushes and shortens the time to completion, and with it the perceived value of the work. What’s left, and it’s almost a cliché now, is judgment.

Shawn: I was part of a fireside discussion a few weeks ago alongside a very talented lawyer who used a couple of terms that inspired this. One is that what now gets delivered is a judgment sandwich. If you’re employing AI, the output is only as good as the direction you give it: the harness, the structure, the prompt, the planning. And on the tail end, you’re left to determine, is this a suitable outcome? Do I take the work product and go with it? What’s in between is now largely commoditized. The other thing he said, specific to legal but broadly applicable, is that much of this has been white-collar manufacturing. This is a guy who probably bills at a thousand dollars an hour, which is itself a bit ironic. But what’s left is to break the billable hour down and see what can be automated, what pieces go on that factory floor. The premium is the judgment.

Jonathan: When I read this, it maps to something David C. Baker, myself, and others have pointed to: the difference between strategy and execution. In software I call them different altitudes. There are three basic altitudes. The bottom is support and maintenance. The middle is execution or implementation. The top is strategy or design, the architecture. That top level, where you need all the judgment or taste, is still where value lives. The execution layer is getting nuked. When I first went out on my own and gave fixed prices, I realized I’d been giving away the judgment, the reason I’m better than someone with less experience, at the same $150 an hour I charged for typing semicolons anybody could do.

Shawn: And much of that judgment was up in the discovery phase, early in the project, right?

Jonathan: Exactly. Picking the architecture, the stack, the direction, and making sure it aligns with what the business actually wants. A lot of times they just give you a punch list, and I’d push back: I can do all of that, but why do you want it done? They’re not the expert, which is why they need me. So I’d uncover the business constraints and goals first, and only then decide how to contribute. If you’re billing by the hour, there’s no need to do that. There’s an old consulting joke: the client signs on the dotted line, you start billing, then go find out what they want. I love how you boiled it down to how it’s all hiding in the same hour. There are really valuable things in there and less valuable things hiding in there.

Shawn: Right. It pulls the lid off. There are people who are excellent at context switching, and maybe they do have different kinds of hours. But by and large, even in the loftiest environments where it’s a twelve-hundred-dollar hour, I think it’s hiding a lot.

Jonathan: When you’re working with clients and they see the downward effect on their revenue, we’re going through work faster than we can get new work because of these tools. I joke, why would you buy a faster computer if you bought it by the hour? It costs you money and penalizes you for being faster. So when you’re in front of a room of partners at big law firms, what are they thinking, and how do you turn the light bulb on?

Shawn: Law is interesting. It has historically enjoyed protection from a lot of this downward pressure. Some will respond, we’re a brand firm, a boutique, we have these high-profile partners. That’s a sensitive conversation. You don’t walk in and tell them their baby’s ugly. But it will only protect you for so long. In every sector I point back to Sequoia, the storied Silicon Valley venture firm. I wrote a piece called “Sequoia put a trillion-dollar bounty on your business.” They put out a call for startups going after what is in aggregate a trillion-dollar opportunity. The long and short of it: they’re funding startups to sell the outcomes, the work that professional services firms do. In law, people respond, well, we have Harvey, we have Legora, brilliant firms doing great work. But you and your competitors are all buying the same tools. It’s a race to the bottom. Why would you buy Opus 4.8 instead of 4.7? It burns more tokens and delivers the answer faster.

Shawn: A lot of it comes down to where they are in the life cycle of the business and of their career. One senior lawyer told me he’s evaluating hiring systems engineers instead of junior associates at three hundred thousand a year. My first response was, I didn’t know first-year associates got paid three hundred thousand. But quite shrewdly he’s looking at how to bring in a systems engineer to build the scaffolding, the workflows, the routines to compress and accelerate the muddy middle. So the long answer is: it’s ego, it’s stage and role, and it’s pressure from clients, which varies by sector.

Jonathan: Quick comment on the marketing piece, which addresses a broader AI concept. The extent to which you can validate the output of the LLM defines how useful it actually is. If it outputs code for a website I could have written, I can read it and know if it’s garbage. If someone pukes out a fifty-page contract or redlines an NDA but doesn’t have the expertise to know if it’s hallucinating, it’s kind of useless, because you have to have it reviewed by an expert anyway.

Shawn: Absolutely. In marketing everyone thinks, I’ll know it when I see it. You can iterate endlessly with an infinitely patient, intern-level marketing AI putting out brand guidelines and colors and fonts. If you’re doing Facebook ads, you can be deceived into thinking, that looks good, I’d click on that, I’ll use it.

Jonathan: I suspect there’ll be a pendulum swing back when someone spends ten thousand dollars on Facebook ads they generated themselves that don’t work, and then thinks, maybe I should talk to an expert.

Shawn: You raise an excellent point. Not to get too in the weeds, but there’s a plugin called Marketing Skills by a talented fellow, Corey Haynes, and the impeccable style. I don’t know him, I just know his work. It does a brilliant job of conversion rate optimization, cold email, ads, video. Better in the hands of someone who knows, to your point. I’d pay someone for their pattern recognition, experience, and judgment, in concert with doing some of the grunt work myself, because running a startup is budget-friendly. The question I ask many clients is, what’s the cost of being wrong? The contract versus the Facebook ad that flops. If I don’t respond to a redline properly in a contract, that could be devastating. An ad that flops is a very different outcome.

Jonathan: Exactly. Is there an emerging framework, kind of like the new digital transformation? Now we’re doing AI transformation.

Shawn: One point I call out: if you’re a solo or boutique, you’re nimble and can try things. The big four have collectively put ten billion dollars into AI. If you’re in the middle, that’s where it’s tricky. Where we focus is commercialization. The workshop I run is two things: business model transformation, from Alexander Osterwalder, whom I had the chance to train under back in 2010, and customer discovery and product-market fit. Ultimately it’s scoping, speccing, designing, packaging, and pricing new offerings that you test before you go build. I’m not rolling in to say which tools to pick or how to architect systems, but what will endure and what your clients will value. I look at it as three frameworks and four arcs. One is the delivery shift: episodic to continuous, discrete projects to always-on delivery; reactive to proactive, waiting for the call versus surfacing problems first, like an always-on agent watching for a compliance or brand-fidelity problem. In pricing, hours to outcomes, billing time to billing results. And moving up the productization ladder, from bespoke to scalable, custom work per client to codified platforms. There are no silver bullets, and you don’t get from A to Z immediately, but those are the spectrums that frame the work.

Jonathan: That reminds me of a video on Lenny’s podcast, an interview with Benedict Evans, a tech analyst since the mobile days.

Shawn: I’ve always been a fan. Long-time subscriber. Great newsletter.

Jonathan: One thing he pointed out: the big four, Bain, McKinsey, should be firing people, but they’re hiring like crazy. He said, you’d think if you’re going to get a hundred and fifty experts for free from Anthropic or OpenAI, why hire more experts? The question is, what’s the hard part? The time-consuming part was creating the seventy-five-page deck. But the hard part is going to the factory, walking the site, interviewing the stakeholders, having the judgment on both ends. Their clients don’t have people sitting around doing nothing, so who’s going to do the integration work? The McKinseys can’t staff up for it instantly. There still has to be a person doing big pieces of this, even if the deck gets generated in three minutes instead of three weeks. And the client’s just jamming the deck into Claude anyway for the TLDR.

Jonathan: So when you’re talking to someone in a situation like that, not a consulting firm but a manufacturer, are they coming at it as, this will decrease our costs by lowering headcount, like the lawyer with the three-hundred-thousand-dollar associate? Or, this will let us deliver a product or service that couldn’t have existed before? What’s the percentage?

Shawn: Great question. Though my focus is professional services, I’m adjacent to other sectors through firms I partner with who do the build, design, and run. What I hear consistently is that in the beginning it’s squarely about cost savings and cost recovery. Stage one is, we receive fifty spreadsheets a week from suppliers and have to ingest all of it. That’s low-hanging fruit. Most haven’t yet thought about what they can sell now that they couldn’t before. Should I say this in public? Maybe I’m early. There’s one firm here in Nashville, Nashville Automation Company, and I know the founders, so disclaimer, but the tagline is something like “make Monday suck less.” They go in and tackle grueling workflows. So most are looking at cost recovery and cost cutting. It’s the leaders who are thinking about what they can offer now that they couldn’t before, or couldn’t profitably. I’m also a big fan of a company out of New York called Every.to. They’re a twenty-five-person firm where everybody gets an agent. I often say, only somewhat tongue in cheek, what would you do with another ten employees? Or, in the case of an agent, a hyperactive, extremely smart, slightly less focused intern. Where it’s headed is not so much whether we lay off, but what we can do now with everyone having a chief of staff sitting beside them.

Jonathan: It’s funny you said chief of staff, because you also said intern. The mental model, maybe more so with soloists, seems to be people trying to outsource the thinking, the high-level stuff they’re actually good at, to an LLM, versus the tedious administrative stuff no one wants to do. I use the intern metaphor all the time: an entry-level employee who needs handholding, from a security standpoint and a trust-the-output standpoint, and who sometimes nods off, like the server falling over. It needs babysitting, but a lot less than collating all those inventory reports.

Jonathan: It’s a failure mode for a high-level strategic thinker to try to outsource high-level strategy to these models, because they’re trained on such a large pile of data that they regress to the mean. Strategy is almost not allowed to do that. You have to come up with an approach to winning by applying your strengths to weaknesses, something no one else would have committed to. By definition it has to be novel, which you’re not going to find by summarizing the internet. To put a bow on it, I don’t actually know what a chief of staff does, but it sounds executive-level. If it’s an executive assistant, the metaphor works. If it’s someone applying a lot of judgment with autonomy, I’m not sure it’s my favorite metaphor, but I’ve heard a lot of people use it.

Shawn: I should be clear. On the spectrum, we’re closer to intern. The promise is closer to chief of staff: a more trusted confidant with more leeway, authority, budget, and resources. In my own experience, it’s all about the scaffolding. The more time I invest in skills and plugins, the way I found Every.to was through their compound engineering plugin, the better. With artifacts like product.md and design.md, shout-out to those who’ve been using Markdown for fifteen or twenty years, it’s paying off. Those seemingly innocuous artifacts are the guidelines. They’re not infallible, but they let you bake more of the judgment in. The more I invest in that for certain projects, the more predictable the outcomes, and I now have processes running overnight.

Jonathan: Let’s drill into it, because people listening probably want to know. I’m fairly deep into this space.

Shawn: It’s possible to have very specific, focused agents that run as a GitHub workflow, an action, doing targeted things, and in aggregate you get pretty great outcomes.

Jonathan: One thing I’ll point out to people who are surprised there’s so much talk about AI that isn’t reflected in their own experience: when people say AI, they’re talking about wildly different things. Let me give you three quick level-ups. First, if you’re not paying for an account at all, it’s probably not that great; once you pay, it gets better right away. Second, it doesn’t have much context about you if you don’t use it much, but if you use it a lot, it has tons of context. If you’re someone like me with ten thousand markdown files from ten years of writing, and you give it access, it can produce incredible things, like a gap analysis of what I haven’t written about that I talk about all the time. Third, when it can write files, move things around, and interact with APIs on your behalf, that’s another huge level-up. I can say, take all of these files, rename them in reverse date order, and upload them to my Transistor site, and it’s just done. That’s wildly different from asking ChatGPT to write a book report.

Shawn: That’s the difference between a one-shot and a workflow. When we wrap and you send me the link, I’ll drop into Claude Code, use a slash skill, give it the link, and it builds the page top to bottom, including the transcript and the overlay that blocks the preplay. Why? Because I like things to look the way I like them. I don’t want to move a div tag; I gave that up long ago, but I can point a single skill at a single link. I do the same for research. I’ll take the latest McKinsey report, use one skill, and it goes into a thesis file and flags me when something runs against the thesis. I have hypotheses about how things will play out, and I want to be challenged and to sharpen the tools. It’s a whole other level than jumping into ChatGPT, which most people use as a replacement for Google search.

Jonathan: Exactly. There are whole other worlds in people’s minds when they talk about this. It’s more like having an employee. It feels exactly like delegating, not like automating. So where do people go next? I’m curious about bigger firms. When a company that size goes through a transformation like this, it makes sense that ninety-five percent would sell it internally as cost reduction or recovery. So what do they do, hire you or someone like you to come in, essentially a consulting gig with this AI-transformation concept, and deliver one working proof of concept to demonstrate the savings, then build from there?

Shawn: I wouldn’t advise against that. Hands on keyboard is the new equivalent: moving beyond ChatGPT into a terminal, using Claude Code or Codex, within the confines of what you allow it to access on your file system. The trick is, say you’ve got thirty employees. I’m working with a financial services firm of about a hundred and twenty, and thirty or forty of them have various tools and are seeing benefits. But because of the complexities of financial services, they’re not yet sharing that product.md or design.md, not submitting PRs to a project. They’re all experimenting individually, which blindfolds your sense of what’s possible, because only so much gets done on one person’s machine. So first get hands on with individual experimentation, then within a function or department pick a project. Shameless plug: at upshiftco.com I have an assessment that measures how exposed a firm is to the billable hour, and where the activity is, and whether it requires senior involvement. If you’ve got exposure and it’s a lot of your work but doesn’t take senior people, pick that and see what’s possible. What I deliver with Upshift is a two-day engagement, a one-day workshop: get the senior partners in the room, work through service design and pricing, and at the end deliver two to three offering briefs. Those briefs are packaging, pricing, positioning, the first three moves, and who among your existing clients to test with. Go test it, see if it’s a fit, and then build.

Jonathan: I’m curious about the state of affairs when you enter one of these companies. Here’s what I think you’re implying: leadership said everybody gets a Claude account or an OpenAI account, here’s your budget, play with it. So individual silos of expertise, growing or not. Is that more or less right?

Shawn: It is, and it’s twofold. Part is positive: let’s get familiarity, get hands on. The other is often a response to an uproar from employees saying we’re at a disadvantage because we’re not using these tools.

Jonathan: Really? Not the reverse, employees saying they hate it, it’s killing the planet, they don’t want it?

Shawn: There are those too. If you go into the design function, you may hear, get the slop away from me. If you’re in a go-to-market function where you need to source leads, enrich data, and automate outbound, it’s phenomenal. The other piece is a lot of “we need an AI strategy,” which is a faster version of “we need a mobile strategy, a cloud strategy.” The friction is, we’re on the back foot, we have to do something, or there’s an uproar from the employee base. My position, not always the most popular, is to slow down, step back, and think about how you’re actually going to commercialize this, not just cut costs.

Jonathan: Exactly. Well, this has been amazing. In case you didn’t see it, on the cover of the Financial Times a few days ago there was a big article about how AI was killing the billable hour, citing sources from large legal firms. That’s about as mainstream as it gets.

Shawn: Thank you for that. I’ll go find it.

Jonathan: Why don’t you give everyone that link again so they can check out the assessment or forward it to someone in a bigger firm who might need your help.

Shawn: Absolutely. It’s upshiftco.com. Upshift is the business name, upshiftco.com is the URL.

Jonathan: Excellent. Thanks again.

Shawn: Thank you, Jonathan. Pleasure.

Auto-transcribed from the episode audio and lightly cleaned. May contain errors.