Latest / Uphoff on Media Podcast / Production is Getting Cheaper. Judgement Isn’t

Transcript

Production is Getting Cheaper. Judgement Isn’t · Uphoff on Media Podcast

• 0:01
  SPEAKER_00: Hey, it's Tony.
• 0:03
  SPEAKER_00: Last month, I ran about four weeks of back office work through my AI agent.
• 0:08
  SPEAKER_00: It took three days on the calendar and nine hours of my time,
• 0:11
  SPEAKER_00: and the hours that mattered weren't spent producing anything.
• 0:16
  SPEAKER_00: That project is today's case study, and it's on my own business, not a client's.
• 0:20
  SPEAKER_00: Regular listeners know I've been arguing that Agentic AI is about
• 0:24
  SPEAKER_00: to let services businesses run With software scale and software margins.
• 0:29
  SPEAKER_00: Today I'm testing that argument against my own numbers,
• 0:31
  SPEAKER_00: including the parts that don't flatter me.
• 0:34
  SPEAKER_00: I'll also walk through the two business models I think will lead this shift.
• 0:38
  SPEAKER_00: Solo scale and team scale.
• 0:40
  SPEAKER_00: What defines them and what the economics look like.
• 0:43
  SPEAKER_00: As always,
• 0:43
  SPEAKER_00: the companion post on Uphoff
• 0:45
  SPEAKER_00: on Media has two Uphoff advisory frameworks you can download and use
• 0:49
  SPEAKER_00: in your business.
• 0:51
  SPEAKER_00: One is SoloScale and TeamScale, which shows you where your business sits today.
• 0:56
  SPEAKER_00: The other is the judgment lever,
• 0:58
  SPEAKER_00: which helps you decide what to hand to an agent,
• 1:00
  SPEAKER_00: what to keep, and how to manage the handoffs between the two.
• 1:04
  SPEAKER_00: Okay, let's get to work.
• 1:09
  SPEAKER_00: Nine months after launching Uphoff Advisory and Uphoff on Media, I had a discovery problem.
• 1:14
  SPEAKER_00: 74 posts,
• 1:15
  SPEAKER_00: a growing framework library,
• 1:17
  SPEAKER_00: and a suite of client tools,
• 1:19
  SPEAKER_00: and not enough of it was being found in search or recommended by AI.
• 1:23
  SPEAKER_00: Here's the admission.
• 1:24
  SPEAKER_00: I'd spent 20 plus years running companies with teams who did this work.
• 1:29
  SPEAKER_00: I knew what good looked like, but I didn't know how to do it.
• 1:32
  SPEAKER_00: So I handed it to my agents.
• 1:34
  SPEAKER_00: This was my own back office, not client work.
• 1:38
  SPEAKER_00: They built or rebuilt 27 framework pages,
• 1:41
  SPEAKER_00: re-rendered about 30 charts, and audited all 74 posts for broken links.
• 1:47
  SPEAKER_00: Working alone, a skilled person would need about four weeks to do that.
• 1:51
  SPEAKER_00: I spent nine hours.
• 1:53
  SPEAKER_00: Now, it's tempting to say AI did the work.
• 1:56
  SPEAKER_00: It didn't, and a lot of operators are building their plans on that mistake.
• 2:00
  SPEAKER_00: The agent didn't take me out of the work.
• 2:02
  SPEAKER_00: It moved me from production to judgment.
• 2:05
  SPEAKER_00: The leverage came from that shift, not from the automation.
• 2:10
  SPEAKER_00: Soloscale was a thesis, and the data is starting to come in.
• 2:14
  SPEAKER_00: In July,
• 2:15
  SPEAKER_00: I argued in Soloscale that agentic AI breaks the constraint
• 2:19
  SPEAKER_00: that has kept every services business at services margins,
• 2:23
  SPEAKER_00: one expert, finite hours, revenue that scales with headcount.
• 2:27
  SPEAKER_00: I wrote that the operator sets strategy,
• 2:30
  SPEAKER_00: reviews output, exercises judgment, while agents handle the production.
• 2:34
  SPEAKER_00: That was an argument.
• 2:36
  SPEAKER_00: This is a test of it against my own business with the limits stated up front.
• 2:40
  SPEAKER_00: Look, it's one project and one operator.
• 2:43
  SPEAKER_00: The counts of pages, posts, links, and URLs are exact.
• 2:47
  SPEAKER_00: The hours are estimates, not timesheets.
• 2:50
  SPEAKER_00: And nine is the high end of mine.
• 2:52
  SPEAKER_00: Read them as an order of magnitude, not an exact count.
• 2:56
  SPEAKER_00: The four week figure is a line item estimate, not a guess.
• 2:59
  SPEAKER_00: Writing 27 pages is the biggest piece at 40 to 68 hours,
• 3:03
  SPEAKER_00: followed by building them in WordPress and re-rendering the charts.
• 3:07
  SPEAKER_00: Four weeks against nine hours is roughly 18 to one.
• 3:11
  SPEAKER_00: At the low end of the estimate, 125 hours, it's still 14 to one.
• 3:17
  SPEAKER_00: Where my hours actually went.
• 3:19
  SPEAKER_00: The division of labor was clean on paper.
• 3:22
  SPEAKER_00: Decisions.
• 3:23
  SPEAKER_00: I made every published call.
• 3:25
  SPEAKER_00: Nine rounds of three pages,
• 3:27
  SPEAKER_00: plus the names, the standards, and which version of a chart was accurate.
• 3:30
  SPEAKER_00: The agent proposed each one with a recommendation and the tradeoffs.
• 3:36
  SPEAKER_00: Framework pages.
• 3:37
  SPEAKER_00: For the library on my website,
• 3:38
  SPEAKER_00: the agent wrote the pages describing the frameworks I created,
• 3:42
  SPEAKER_00: including how-to steps, FAQs, SEO fields, and alt text.
• 3:47
  SPEAKER_00: It built every page in WordPress and verified each one live.
• 3:50
  SPEAKER_00: I reviewed.
• 3:52
  SPEAKER_00: Charts, the agent re-rendered and rebuilt about 30 and checked the stats against the sources.
• 3:57
  SPEAKER_00: I uploaded them.
• 3:59
  SPEAKER_00: Substack, the agent wrote every link line and alt text and verified every post afterward.
• 4:05
  SPEAKER_00: I made the edits in 22 posts and fixed six links.
• 4:09
  SPEAKER_00: Search, the agent got all 27 pages to green in Yoast, audited 74 posts and checked 31 URLs.
• 4:17
  SPEAKER_00: I submitted the sitemap.
• 4:19
  SPEAKER_00: Records.
• 4:20
  SPEAKER_00: The agent kept the upload, workbook, link map, tracker, and status doc current.
• 4:25
  SPEAKER_00: I didn't touch them.
• 4:27
  SPEAKER_00: Look at what I actually did on that list.
• 4:28
  SPEAKER_00: About half of my nine hours went to Substack edits and another hour or two
• 4:33
  SPEAKER_00: to uploading images.
• 4:34
  SPEAKER_00: That's production, not judgment.
• 4:36
  SPEAKER_00: I was doing it by hand because Substack blocks the agent from that work
• 4:40
  SPEAKER_00: and getting files into WordPress's media library stayed manual.
• 4:44
  SPEAKER_00: So the division isn't clean yet.
• 4:46
  SPEAKER_00: Only two to four of my nine hours went to real judgment calls.
• 4:50
  SPEAKER_00: The rest went to manual work.
• 4:51
  SPEAKER_00: The tools wouldn't let the agent touch.
• 4:54
  SPEAKER_00: That's plumbing and plumbing gets fixed.
• 4:56
  SPEAKER_00: None of it really needs me.
• 4:59
  SPEAKER_00: The checking was worth as much as the speed.
• 5:01
  SPEAKER_00: The number that will get quoted is the hours.
• 5:03
  SPEAKER_00: What the checking found mattered just as much.
• 5:06
  SPEAKER_00: Four published facts were wrong.
• 5:09
  SPEAKER_00: All four were mine, and they'd been live.
• 5:11
  SPEAKER_00: A stat credited to the wrong firm,
• 5:13
  SPEAKER_00: 94% of B2B buyers use LLMs, was attributed to Forrester and CMI on two charts.
• 5:20
  SPEAKER_00: It actually comes from the Sixth Sense 2025 Buyer Experience Report.
• 5:25
  SPEAKER_00: A chart that contradicted its own post,
• 5:27
  SPEAKER_00: when charts said mid-market B2B was finally off-premise in 2018,
• 5:32
  SPEAKER_00: the post it ran, WIPP said the opposite.
• 5:35
  SPEAKER_00: A chart that didn't match its text.
• 5:37
  SPEAKER_00: The post described seven questions.
• 5:39
  SPEAKER_00: The chart showed six.
• 5:40
  SPEAKER_00: A count that didn't add up.
• 5:42
  SPEAKER_00: One framework page said four squares and listed five.
• 5:46
  SPEAKER_00: Look, none of these were catastrophic.
• 5:47
  SPEAKER_00: Each one is exactly the kind of error a careful reader finds, though.
• 5:51
  SPEAKER_00: And the readers of this publication are careful readers.
• 5:55
  SPEAKER_00: I would not have audited 74 posts by hand.
• 5:58
  SPEAKER_00: Nobody running a business has the time.
• 6:00
  SPEAKER_00: The agent found the errors.
• 6:02
  SPEAKER_00: It didn't decide what to do with them.
• 6:05
  SPEAKER_00: The chart that contradicted its post needed a new label.
• 6:08
  SPEAKER_00: That was true in both places.
• 6:10
  SPEAKER_00: The seven-question index needed its scoring range rebuilt.
• 6:13
  SPEAKER_00: Those were calls, and they were mine.
• 6:15
  SPEAKER_00: Verification moved to the agent.
• 6:17
  SPEAKER_00: Deciding what's true stayed with me.
• 6:21
  SPEAKER_00: Judgment becomes the lever, not the cost.
• 6:25
  SPEAKER_00: The build was fast, not frictionless, and the friction is the most useful part of the story.
• 6:30
  SPEAKER_00: The one hour I'd take back came from a call I made late.
• 6:34
  SPEAKER_00: The Uphoff value index went from value to valuation and back again inside of an hour.
• 6:40
  SPEAKER_00: Each switch meant a chart re-render, a page re-title, a URL change, and a redirect.
• 6:46
  SPEAKER_00: Four pieces of work each done twice.
• 6:49
  SPEAKER_00: That waste was mine, not the agent's.
• 6:51
  SPEAKER_00: Agents multiply whatever direction you give them.
• 6:55
  SPEAKER_00: Make the call early and they multiply the work.
• 6:57
  SPEAKER_00: Make it late and they multiply the rework.
• 7:01
  SPEAKER_00: Not every mistake was a decision and not all of them were the agents.
• 7:05
  SPEAKER_00: One round of uploads landed on the development site instead of the live one.
• 7:08
  SPEAKER_00: A stray bulk upload pulled in 29 files.
• 7:12
  SPEAKER_00: All 18 rebuilt pages went to the trash once and came back with undo.
• 7:17
  SPEAKER_00: Every one of those was fixed in minutes because the records were current and
• 7:21
  SPEAKER_00: the standards were already written down.
• 7:23
  SPEAKER_00: That's the operating lesson here.
• 7:25
  SPEAKER_00: In every transformation I was brought in to lead,
• 7:28
  SPEAKER_00: execution was where things broke and the worst breaks traced back to decisions
• 7:33
  SPEAKER_00: that came after the work had started.
• 7:36
  SPEAKER_00: Agents raised the stakes.
• 7:37
  SPEAKER_00: They execute so fast that a clear call made up front pays off almost immediately and
• 7:43
  SPEAKER_00: a late one throws away work that's already done.
• 7:46
  SPEAKER_00: The leverage is in the calls you make before the work starts.
• 7:55
  SPEAKER_00: Last week,
• 7:56
  SPEAKER_00: I wrote about the difference between an output multiple and how much more
• 8:00
  SPEAKER_00: a person produces on one workflow and a scale multiple,
• 8:04
  SPEAKER_00: how much more revenue each person brings in.
• 8:07
  SPEAKER_00: This build is an output multiple, a large one.
• 8:10
  SPEAKER_00: It's not yet a scale multiple, and I'd be overselling it to call it one.
• 8:14
  SPEAKER_00: For anyone new to the terms,
• 8:16
  SPEAKER_00: SoloScale is one expert running agents from say $250,000 to $5 million in revenue at
• 8:23
  SPEAKER_00: a 40% or better EBITDA.
• 8:26
  SPEAKER_00: TeamScale is the next rung,
• 8:27
  SPEAKER_00: a small senior team doing the same at $400,000 or more of revenue per employee
• 8:34
  SPEAKER_00: and 25% or better EBITDA, say a business from $2 million to $30 million.
• 8:39
  SPEAKER_00: Both are defined by revenue per person, not headcount.
• 8:42
  SPEAKER_00: For context, the median private SaaS company does about $175,000 of ARR per employee.
• 8:50
  SPEAKER_00: And WPP, the largest agency holding company, does about $130,000 per person.
• 8:56
  SPEAKER_00: This build is evidence on the input side of the math, not the revenue side yet.
• 9:03
  SPEAKER_00: If you're listening rather than reading, here's how to use that chart.
• 9:07
  SPEAKER_00: Picture three columns side by side.
• 9:09
  SPEAKER_00: A traditional services firm, solo scale and team scale in the traditional firm.
• 9:14
  SPEAKER_00: Revenue grows only by adding people and EBITDA averages around 10%.
• 9:18
  SPEAKER_00: To find your column, ignore headcount.
• 9:21
  SPEAKER_00: Take last year's revenue and divide it by your full-time equivalents.
• 9:25
  SPEAKER_00: Count your regular contractors as people because they are.
• 9:28
  SPEAKER_00: Then figure your EBITDA after paying yourself a market salary.
• 9:32
  SPEAKER_00: Founders love to skip that step, don't.
• 9:35
  SPEAKER_00: Those two numbers tell you which column you're in.
• 9:37
  SPEAKER_00: If you're under $130,000 a head,
• 9:39
  SPEAKER_00: you're running a traditional services firm, whatever your pitch deck says.
• 9:43
  SPEAKER_00: If you're between $130,000 and $250,000,
• 9:46
  SPEAKER_00: you're in the gap,
• 9:47
  SPEAKER_00: and the question is how much of your leader's time still goes to production.
• 9:51
  SPEAKER_00: If you're above the entry lines, the question changes.
• 9:54
  SPEAKER_00: It's no longer whether you can get there, it's how high the ceiling goes.
• 9:58
  SPEAKER_00: The floors are defined.
• 9:59
  SPEAKER_00: Nobody has found the ceilings yet.
• 10:01
  SPEAKER_00: If you fund the services business, run the same math on every company you back.
• 10:06
  SPEAKER_00: It's a quick test of whether a business is headed for services margins or software margins.
• 10:19
  SPEAKER_00: Then there's the cost.
• 10:20
  SPEAKER_00: The agents ran on tools I already paid for.
• 10:23
  SPEAKER_00: Four weeks of the same work hired out at a typical freelance SEO rates of $75
• 10:28
  SPEAKER_00: to $100 an hour,
• 10:29
  SPEAKER_00: according to an AREF survey, would have cost roughly $12,000 to $16,000.
• 10:35
  SPEAKER_00: One thing isn't in the comparison.
• 10:37
  SPEAKER_00: My web developer made a configuration change at no charge that let the pages hold real content.
• 10:42
  SPEAKER_00: It doesn't change the order of magnitude, but it belongs on the record.
• 10:47
  SPEAKER_00: The more important point is what the comparison actually is.
• 10:50
  SPEAKER_00: I was never going to hire a writer,
• 10:52
  SPEAKER_00: a designer, and a web producer to rebuild 27 pages of my own IP.
• 10:57
  SPEAKER_00: Without the agent, the library stays half built.
• 11:00
  SPEAKER_00: That's the real solo scale effect.
• 11:02
  SPEAKER_00: The work isn't just cheaper.
• 11:04
  SPEAKER_00: Work that would have never been funded now exists.
• 11:07
  SPEAKER_00: Whether it turns into a scale multiple depends on what the operator does with
• 11:11
  SPEAKER_00: the hours the agents gave back.
• 11:13
  SPEAKER_00: While the agent was building the library, I brought in a new advisory client.
• 11:17
  SPEAKER_00: Look, one additional client isn't a scale multiple,
• 11:20
  SPEAKER_00: but it's exactly where those hours are supposed to go.
• 11:23
  SPEAKER_00: If they go into more client work and better client work, the margin shows up.
• 11:27
  SPEAKER_00: If they go into more production, it doesn't.
• 11:31
  SPEAKER_00: I work more hours, not fewer.
• 11:33
  SPEAKER_00: Here's the part I didn't expect.
• 11:36
  SPEAKER_00: You'd think a solo scale business means working fewer hours.
• 11:40
  SPEAKER_00: I work more.
• 11:41
  SPEAKER_00: What changed is where they go.
• 11:43
  SPEAKER_00: When I launched the firm,
• 11:44
  SPEAKER_00: a real share of my week went to production,
• 11:47
  SPEAKER_00: formatting, uploading, fixing links, keeping records.
• 11:50
  SPEAKER_00: Most of that now sits with the agents.
• 11:53
  SPEAKER_00: Those hours moved to the work clients actually hire me for.
• 11:56
  SPEAKER_00: Research, analysis, planning, and strategy.
• 11:59
  SPEAKER_00: It shows up in the engagements too.
• 12:01
  SPEAKER_00: Less of my time goes to the production that comes with the project work,
• 12:05
  SPEAKER_00: so more of it goes to the connected issues that decide whether a business grows.
• 12:09
  SPEAKER_00: Positioning,
• 12:10
  SPEAKER_00: packaging and pricing,
• 12:12
  SPEAKER_00: the roles a team needs and who fills them,
• 12:14
  SPEAKER_00: channel strategy and coaching the leaders who have to carry it all.
• 12:18
  SPEAKER_00: None of that is production.
• 12:19
  SPEAKER_00: All of it sits in the keep it box.
• 12:22
  SPEAKER_00: That's what a services business with software margins looks like up close.
• 12:26
  SPEAKER_00: It's a judgment business with a production engine underneath it.
• 12:30
  SPEAKER_00: The engine does more every quarter.
• 12:32
  SPEAKER_00: The judgment is what clients hire me for.
• 12:38
  SPEAKER_00: Here's an Uphoff advisory framework you can put to work on your next project.
• 12:42
  SPEAKER_00: I call it the judgment lever.
• 12:45
  SPEAKER_00: Picture a two by two.
• 12:46
  SPEAKER_00: Across the bottom, who does the work?
• 12:48
  SPEAKER_00: The agent on the left, you on the right.
• 12:51
  SPEAKER_00: Up the side, what kind of work it is?
• 12:53
  SPEAKER_00: Production at the bottom, judgment at the top.
• 12:55
  SPEAKER_00: Four boxes.
• 12:56
  SPEAKER_00: Bottom left is handoff.
• 12:58
  SPEAKER_00: Agent, production, drafting, building, rendering, record keeping, the volume.
• 13:02
  SPEAKER_00: Top left is propose, don't decide.
• 13:06
  SPEAKER_00: Agent, judgment, fact checks, audits, options.
• 13:09
  SPEAKER_00: The agent brings you a recommendation and the trade-off.
• 13:12
  SPEAKER_00: You make the call.
• 13:14
  SPEAKER_00: Top right is keep it.
• 13:15
  SPEAKER_00: You, judgment, name, standard, what's true, what ships.
• 13:19
  SPEAKER_00: Decide those first.
• 13:21
  SPEAKER_00: Bottom right is fix the plumbing.
• 13:23
  SPEAKER_00: You, production, manual work where you're still doing because the tools won't connect.
• 13:28
  SPEAKER_00: It should shrink every quarter.
• 13:30
  SPEAKER_00: If it doesn't, start there.
• 13:32
  SPEAKER_00: The move is the diagonal.
• 13:33
  SPEAKER_00: Hand off the bottom left so you can spend your time in the top right.
• 13:38
  SPEAKER_00: The full framework is in the post or at upoffadvisory.com slash frameworks.
• 13:49
  SPEAKER_00: What to do with it this quarter.
• 13:51
  SPEAKER_00: If you're building,
• 13:51
  SPEAKER_00: running,
• 13:52
  SPEAKER_00: funding,
• 13:52
  SPEAKER_00: or starting a services business,
• 13:54
  SPEAKER_00: run one project through the judgment lever and keep score.
• 13:57
  SPEAKER_00: Make the keep it calls before production starts.
• 14:00
  SPEAKER_00: Names, sources, standards, what done looks like.
• 14:04
  SPEAKER_00: Back where your own hours land,
• 14:06
  SPEAKER_00: if most of them sit in the fix the plumbing, you don't have a judgment problem.
• 14:09
  SPEAKER_00: You have an integration problem, and it's the cheapest one you'll ever solve.
• 14:14
  SPEAKER_00: If you fund services businesses, ask the operators you back the same question.
• 14:19
  SPEAKER_00: How much of their leader's time still goes into production?
• 14:23
  SPEAKER_00: The agent will make more of the work We'll take more of the work every quarter.
• 14:28
  SPEAKER_00: It won't make the calls and it can't make them early on your behalf.
• 14:31
  SPEAKER_00: Hand it the production, keep the judgment calls and make them first.
• 14:38
  SPEAKER_00: Well, that's the episode.
• 14:39
  SPEAKER_00: To download the Judgment Lever and Soloscale and TeamScale,
• 14:42
  SPEAKER_00: go to the post on Uphoff on Media on Substack
• 14:45
  SPEAKER_00: or visit uphoffadvisory.com slash frameworks.
• 14:49
  SPEAKER_00: That's uphoff, U-P-H-O-F-F, advisory.com slash frameworks.
• 14:55
  SPEAKER_00: And one favor,
• 14:56
  SPEAKER_00: if the show is useful,
• 14:58
  SPEAKER_00: follow it and leave a rating or review on Apple Podcasts or Spotify.
• 15:02
  SPEAKER_00: It's the easiest way for other operators to find it.
• 15:06
  SPEAKER_00: Back at it next week.