Incremental AI

For every AI skeptic and alarmist, there are ten AI dreamers convinced AI will change civilization. In the business world, this fuels high valuations and premature optimization through layoffs or sweeping business changes.

Amazon's CEO, Andy Jassy, listed out all sorts of AI things Amazon is doing. To me, the list suggests AI is great at incremental improvements to existing business processes.

For instance, improving campaign planning and copywriting helps Amazon and sellers. But if the product still sucks or is overpriced, who cares? Getting a mini-grill delivered on time for the weekend camping trip is nice - but not the singularity.

Incremental AI isn't a grand transformation. If leaders talked this way, you'd hear things like: “We estimate AI-driven improvements will boost productivity by 5% to 10% and cut costs accordingly. We'll run our business more efficiently.” (Oh… and finally deliver on those Alexa and Siri promises from 2014 and 2011)

This kind of incremental improvement is fantastic and justifies a lot of the hype. However, focusing on earth-shattering results sets expectations incorrectly, and then the fall from those expectations is damaging.

We should instead expect boring but good. In that scenario, AI can help with predicting lost luggage, setting prices, basic strategic planning, and even solving the original Jobs to be Done question (how to sell more milkshakes) with enough data.

AI excels at common sense - perfect for organizations that aim to operate steadily, improving gradually. Their core strategy: don't fuck it up.

Generative AI shines at content creation (text, images, videos). If your work involves text, AI will improve it.1 This is why it can enhance programming - possibly the field AI will impact most.

This implies:

1. AI can't reason enough to come up with action to take. It delivers predictable, common-sense outputs—token by token. (Sure, you can adjust the temperature for randomness… but is that what you want?) It has no initiative and, like, “get shit done” attitude.

2. AI excels at working with known events and past thinking (if it was written down), not imagining new things. After 2+ years playing D&D with the big three AIs, I can tell you: AI can't create the unexpected and it can’t take initiative. Productivity with an AI isn’t great. Even with coaching, a human DM is still better and far more efficient.2

3. AI handles "day one" programming - the first commit. From what I know, it’s also great at refactoring and modernization, especially the part where you’re trying to figure out what that ten year old code does and why. Bu, you still need to evolve, architect, and maintain the code. As was pointed out recently, writing code is just a part of “programming.” common-sense focus may make it great for product management: an AI PM might push for adding LDAP support over chasing the latest shiny tech.

Business-bullshit automation

AI's biggest workplace value likely accrues to individuals. It's great at busywork—tasks everyone knows the answer to, but require endless slides and meetings, for example, what should the salty-snacks company do to improve sales in the south west next year: “In FY2027, we should sell introduce unflavored (“natural”) tortilla chips in Texas with a nostalgic grandpa on the bag that mentions they’re fresh made from whole tortillas, not ground up corn reformed into triangles like THE OTHER GUYS. Also, make them blue in New Mexico."

AI also excels at navigating enterprise bureaucracy. Here's an example from the King of Ketchup:

Team members use the platform to analyze and summarize documents or dense industry reports, as well as to optimize and access standard operating procedures on a factory floor.

Are your documents poorly written, confusing, and hard to find? AI don't care! It'll locate the policy for filing factory expense reports over €499 in Friesland.

IRL benefits

AI is already benefiting real life most: therapy, learning, tutoring, cooking, basic medical advice, and home how-tos (a task YouTube has done well for years).

Measuring ROI and “productivity” here is impossible - as with most of real-life. The historic parallel is Internet search, which improved everyday living. There the ROI is kind of infinite, considering that the cost is zero.

Like search, AI could improve your everyday life - until spam and shit-content swamp it. Remember when you could quickly find a tortilla recipe? Used to be you could quickly look-up how to make tortillas, and now you have to first read ten stories of how the author grew up making tortillas with their uncle and this special bond moved them into the propane tank industry where they first helped their uncle at the local gas station and, over 20 years, grew a $5bn direct-to-consumer business.

The CEO Take-out Test

Like I said above, AI lacks imagination and, I don’t know, “gumption”: making a decision about what to do and then doing it. Perhaps it’ll get good at it, and then that would be more than incremental improvement. A simple test for that happening: could AI replace a CEO? A board? If yes, it can likely reason, act, and imagine.

Until then, AI delivers incremental improvements, enterprise bullshit automation, pictures, video (Gemini is amazing lately!), and programming.

Relative to your interests

• What’s New in VMware Cloud Foundation 9.0 - huge release from our big sister division at Broadcom. Highly related in Europe:

• ‘We’re done with Teams’: German state hits uninstall on Microsoft - You’re going to see a lot of Europeans do this kind of thing. I’ve already had multiple conversations about “cloud patriation,” let’s call it.

• Why is digital sovereignty important right now?

• Digital News Report 2025, Reuters and Oxford - “These shows are filmed, with many people watching on YouTube - the most popular platform for podcasts in the country, with 50% of US podcast consumers using it.” // Also, New York Times and Joe Rogan very popular in the US. // And, demographics: “The report examined survey responses from roughly 97,000 adults across more than 45 countries.” // “They found that 7% of Americans overall - and 12% of those under the age of 35 - use AI platforms for news.”

• Making the Invisible Hand Visible: Managers and the Allocation of Workers to Jobs - Match workers to jobs they’re good at. // “My results imply that the visible hands of managers match workers' specific skills to specialized jobs, leading to an improvement in the productivity of existing workers that outlasts the managers' time at the firm.”

• ‘But I’m not creative’: The role of creativity and connection in content design - An overlooked part of creating: establish your taste in things and apply it to your work. When working with (editing) others, figuring out what their taste is, and apply it to their work.

• Will Zero-Click Search Kill My B2B Website? - Make sure your marketing content gets picked up by the AI. This probably means the standard SEO stuff: put out a lot of it loaded with keywords and rephrasing, spread across different URLs. Related:

• Your Next Move: Capitalize on Marketing Your AI - “As AI pushes this evolution even further, 70% of U.S. B2B buyers will rely on GenAI tools throughout their buying process by 2028.” // Advice on marketing AI stuff.

• Edisum: Summarizing and Explaining Wikipedia Edits at Scale - A prompt for summarizing edits made to text.

• UX Case Study: SubStack WIP - UI tricks.

• Detecting AI-Generated Text by Uncovering Its Statistical “Tells”Cloud - Community - ”The next time you read something that feels just a little too perfect, pause. You might not just be spotting an AI. You might be feeling the absence of the beautiful, chaotic, and unmistakable fingerprint of a human mind.” // You know, I would treat this kind of thing as an overview of the robot’s default style. You can train it to write like you (or whoever) and remove a lot of these “tells.” But, that training takes time and refining, so most people will likely just go with the default, uh, “AI voice.”

• How Kraft Heinz measures AI project value - ”Team members use the platform to analyze and summarize documents or dense industry reports, as well as to optimize and access standard operating procedures on a factory floor. The engine also helped streamline SAP rollouts in China and automate routine financials. Nestor said he has the platform bookmarked in his browser and uses it on a daily basis. ”

• INFILTRATE. SURVEY. PERCEIVE by Reyes Makes Games - This is fun, kind of pranky game to play in a house with others.

• Speaking of tortillas…

Wastebook

• ”Suffice to say, 90s were great so I approve.” American abroad data-points

• And: “the restaurant (in Bordeaux) was out of Bordeaux but said they gave a great New England IPA, and the bar in a part of town with no tourists was playing old NFL highlights.”

• “Fucking only stall.” Stay classy, New York.

Logoff

The SpringOne conference is coming up soon, August 25th to 28th. If you’re one of the millions of people who do enterprise software with Java, you should come check it out.