At a recent lunch out with neighbors, the conversation turned to local restaurants. Some of the places shared were familiar, other suggestions (or cautionary tales) were ones we’d heard of but hadn’t tried. I came away from that chat with a list of places to try, revisit and avoid.
That sparked the idea for this Spring Green Insights post: we’re sharing some of our recent AI tool “tastes” — the use cases, if they worked well, fell short or left us undecided. We’ve been exploring how best to make AI work for us, while ensuring our team stays front and center in the process, as cautioned by this article from Chief Executive.
Dig in!
-Tia
KATHERINE ON CHATGPT:
I use ChatGPT as a thinking partner. It helps me pressure-test ideas, organize complex information and get to a clearer point of view faster.
After SME interviews or internal working sessions, I’ll use rough notes or transcripts to identify themes, surface patterns and structure the narrative. That helps me walk into client conversations with sharper thinking and fewer gaps.
I also use it during concepting. Instead of starting with a blank page, I use ChatGPT to expand ideas, explore directions and challenge assumptions before narrowing in on the strongest path forward.
That said, it’s not a finished product. It doesn’t know the client, the nuance or the stakes. Every output needs judgment, editing and a healthy level of skepticism.
TOP BENEFIT: Faster clarity and stronger early-stage thinking.
DOWNSIDE: It can sound confident even when it’s wrong, you still have to do the thinking.
NATE ON CLAUDE:
Using AI models for code generation is currently where AI shows the most promise. You still need to quality-check and supervise results, but for quick coding tasks, you simply cannot work at the speed these models allow.
In a recent client presentation, for example, there was a need for an animated counter going from 100,000 to 1,000,000. Claude was able to generate the animation first as HTML and then, using Python, produce an animated GIF from that output. In a recent website production completed at Spring Green, numerous short HTML sections were needed, and I could quickly upload a design to get back usable code. Because code is text-based and governed by consistent rules, it aligns well with how current large language models work.
TOP BENEFIT: Rapid code generation.
DOWNSIDE: The models can produce plausible-looking code that contains logical errors or security issues, so the idea that someone can quickly generate production ready software is still some time away.
KATHERINE ON JASPER.AI:
We used Jasper AI as part of a research-heavy project where we needed to gather data on about 15 individuals and develop profiles, along with benchmarking peer companies to inform a client strategy.
Jasper did a solid job on the initial data scrape. Though it struggled with multiple asks in one prompt. It wanted to do each step with a separate prompt. Even with that, which was frustrating at times, it helped us move more quickly than we could have using traditional research tactics.
Once we had that data, the humans had to take over to verify every output, refine it and put it into context. We also ran the same prompts through multiple tools, including Perplexity AI and ChatGPT, to compare outputs and catch gaps or inconsistencies. That extra step improved accuracy and gave us more confidence in the final direction.
The biggest takeaway is that no single tool gets you all the way there. Jasper, like the others, works best as part of a team. It speeds up the process and expands the field, but judgment, context and strategy still sit with us.
TOP BENEFIT: Fast starting point for research and data gathering.
DOWNSIDE: It takes some babysitting to break a complex task down into step-by-step asks and enter each prompt to get the information.
TIA ON COPILOT:
After our Spring Green retreat, we had a bunch of notes on flipcharts. I took photos from my phone, uploaded them to the Copilot app with a prompt to prepare notes. In some cases it was a table, in other cases a list ordered a certain way. It was fast and easy.
A more strategic way I’ve used Copilot was to bring a feasibility study we recently completed for clients into an outline for PowerPoint slides. I vetted that outline and notes Word doc and prompted PowerPoint to build a deck for a board meeting. I still spent a fair amount of time proofing, editing and manipulating charts and AI images (misspellings/disjointed icons/infographics & charts that misrepresented the data) but it was superior to starting from a blank canvas.
TOP BENEFIT: Time savings, integration with our Microsoft 365 system.
DOWNSIDE: Errors small and large don’t build trust. It’s a launching pad, NOT a finished product.
