as seen on Little Black Book Online (lbbonline.com)

The VP of engineering at H/L on the integration of AI into workflows, recent success stories, and meta-prompting, as part of LBB’s AI Spy series

 

John Jamison is VP of engineering at H/L, an independent agency where he’s building the systems behind multi-million dollar marketing campaigns, which is a fancy way of saying he’s the guy responsible for making sure the robots behave.

He built H/L’s core data platform, and now he’s wiring the agency for agentic AI, creating flows and processes to do things the industry’s never been capable of before. It’s automation, pushed to make work faster, more consistent, and higher-quality, on the premise that AI doesn’t replace expertise. Instead, it rips the training wheels off it.

John sat down with LBB to discuss AI-assisted storytelling, the delicate balance between IT, security, and bureaucracy, as well as AI video generation.

 

LBB> What is the most impactful, specific task or process that AI has permanently changed in your current role, and how does it drive qualitative improvements (not just efficiency) to your final output?

John> The thing that AI has changed the most in my particular line of work is storytelling. Handling data and thinking about the narrative around it used to be a monolithic challenge, and once you get to a certain scale, it becomes exceedingly difficult to find the right mix of folks that can explore, clean, and curate interesting angles.

By letting robots take the reins on exploration and cleanup, it has allowed us to focus much more on the story and narrative we present to clients. AI has significantly altered even the way we show up, as most folks are now producing custom-branded HTML pages to present things like dashboards and slide decks.

The velocity created by these advancements has allowed more ideas to come to the surface, and it’s allowed those ideas to come from a wider pool of people at the agency.

 

LBB> AI output is only as good as the input. What is the single most valuable lesson you’ve learned about prompt engineering or interaction design to get truly distinctive and high-quality creative work from an AI model?

John> You need a ‘long-term’ memory for the robot. There is an old developer maxim called ‘DRY’ (do not repeat yourself). This is especially true in the age of AI, where most people exhaust themselves in managing context, regurgitating, copying, and pasting the same pieces of information all the time. You don’t need to get super fancy with a vector database – having some wiki software like Notion or Outline wired up via MCP allows everyone in your workplace to share from the same pool of tribal knowledge.

Tribal knowledge is the (generally) unwritten operational stuff that lives in people’s heads – the stuff you generally learn only by being there. We have made it our goal to surface as much of that detail in a structured way via wiki. This has had tremendous gains for much of our company, as it lets us distribute important work much more effectively. When someone asks a question to their bot or is trying to produce a work product, they don’t all need to flock to the same four people who have been here forever. They are able to get the precise information they need or have a robot search for it just by asking.

 

LBB> The ethics of using AI are centered around data sourcing and IP. What is the most critical ethical guardrail or internal policy your organization has implemented to ensure your use of AI to generate or assist content is responsible and legally sound?

John> This is a nebulous one. Our number one policy is that nothing major to the clients goes out without review. We have a number of policies around AI, but people don’t generally sit around remembering all the legal jargon they’ve read. We try to keep the heuristics simple for the individuals.

You are probably in bad territory if:

• Whatever you have produced could be replicated almost immediately by the kids of your client.
• There is nothing of ‘you’ in the work.

Number one is pretty self-explanatory, while for number two, we care about people being able to pour their own thoughts, expertise, and personal flourishes into whatever they produce. AI generally creates a great ‘averaging’ of everything it’s seen, so to make it original and interesting, you really do need to pour yourself and your eccentricities into it.

 

LBB> What is the biggest challenge in collaborating with AI as a creative professional? Specifically, how do you work to ensure that the final work maintains a demonstrable sense of authenticity, human intuition, or your signature creative identity – and how do you personally avoid getting sucked in to dependence, ‘work slop’ mode or switching your critical brain off?

John> There is a weird mentality shared by LinkedIn posters and a certain class of creative SASS companies where they think the creative process should be a ‘factory’. That mentality is great for system development and engineering, but not really for the creative process. We are the types that think quality over quantity.

AI is a great tool; it is letting us do things we have never done before, and that is the important quality of it that we should chase. So, our approach is a hybrid one. We still bring the same creative process, tools, and techniques that we did before, but now we find ourselves asking questions we wouldn’t have before.

One example is from a recent shoot for our Toyota client. We were drifting a vehicle through some snow and thought it would be sick if we could do a bullet-time-style shot around the vehicle. After wiring up six cameras and getting the staging shots we needed, we were able to generate the missing footage in the rotation between the cameras to get something that was really slick and worthy of Keanu in post.

 

LBB> Have you found that the integration of AI tools (e.g., LLMs, image generators) creates new points of friction or collaboration challenges with other creative or production partners? If so, how are you overcoming them?

John> There is certainly a blending of roles, and this has created a bit of friction in the organisation, as I am sure it has for many other agencies. I am of the school of thought that it is often better to beg for forgiveness than ask for permission. It’s never going to be perfect, but a culture of permission is something that we try to foster at H/L.

Most of the time, it is about trying to bind folks with as few rules as possible. Part of that is stewardship; you can only minimise the rules if everyone is playing as one team. Personal responsibility and collaboration are important in an environment where the number cruncher could be helping with video generation the next day.

 

LBB> As AI rapidly changes workflows, the skillsets required are evolving. What is one new skill or capability related to AI usage that you believe is essential for people in your discipline to learn in the next year?

John> Two big things: meta-prompting and skipping the keyboard. These were the two biggest unlocks for me and the team.

Meta-prompting is this concept that basically says that unless you are an expert in that given field, you probably don’t know the best way to get to your desired outcome. To overcome this, you turn the machine on itself and ask things like:

• What would you need to know to execute this goal effectively?
• What materials or information would help you arrive at this desired outcome?
• I am trying to achieve X, how would you formulate the prompt?

This often gives you language, concepts, and directions that you would have never thought to use when trying to direct the machine.

Secondly, we are now in a world of incredible real-time transcription software. I personally use something called Wisprflow. If you type 60 words per minute or less, your output effectively doubles by just talking at the computer.

 

LBB> There’s some evidence that the productivity gains of AI have been somewhat overstated at an organisational level (or perhaps businesses are still in the discovery phase). How are you making sure that you’re using AI in a genuinely effective and impactful way and are there any areas that you’ve discovered that AI tools can have a detrimental impact on productivity/effectiveness?

John> I actually don’t think that they’re overstated; the problem is that there are some people running at an all-out sprint while others haven’t even crossed the starting line. I’ve had a token burn in the last 30 days of about five billion tokens. That’s roughly 50,000 copies of a ‘Harry Potter’-sized novel in terms of text. I’m an exceptional user. In most organisations, these types of people have already been identified. You know your AI power users generally, and you know that they are transformative for the business.

The problem is that getting good with AI and provisioning others to get good with AI are two radically different problems. I don’t think twice about wiring up different types of software, writing scripts, or diving into the code to get an AI system that is customised to my usage and use cases.

That isn’t the experience of the average user; you need to find ways to unlock big tranches of capability to the widest amount of people possible. This is a delicate balance between IT, security, and bureaucracy. Every organisation is different. You need to figure out if you are going to focus on true believers or the masses, and if you focus on the masses, how you are going to provision and train them to be effective.

 

LBB> Do you think AI is currently creating entirely new forms of art or media that weren’t possible before?

John> Absolutely, video generation is getting really interesting, especially with things like Seed Dance. It still has imperfections, oddities, and sometimes strange proportions, but I think when you’re really trying to produce something interesting that has genuine intention, people will actually often forgive those eccentricities.

That’s been my experience with doing silly things like remaking the intro to ‘Power Rangers’ with all my direct reports. Again, I think the biggest friction point is when people try to take the creative process and convert that into a factory. When all you’re doing is pumping out a hundred videos of pretty influencers using the same tired three voices out of ElevenLabs to tell you about Blue Chew, you can see why we get the ire of the creative industry.

 

LBB> Thinking about your own role and discipline, what kind of specific, measurable impact do you think AI will have in the medium-term (3-5 years)? To what extent will it change the way people in your role work?

John> While I’m most excited about the creative end of things, it will definitely be the most transformative in data and storytelling. Especially in the less sexy side of data, which is really all the cleaning and munging and bullshit that nobody really wants to look at.

Machine lets you do that at scale in plain language, with huge speed gains for most users. It’s a double-edged sword. You will definitely be able to tell the difference between an organisation that is sophisticated in AI and one that isn’t, based on the data and outputs from the data that you receive.