Our take: our rec is Pathway B, Gemini as the base for everyone plus one specialist tool for the heavier synthesis work. This page lays out all three pathways with their costs, a risk comparison, and how we'll settle the open questions.
DRAFT V1
The session happened and the choice is made: Pathway C, with Claude as the second tool, the whole team on governed seats alongside Gemini. One correction since this pre-read was written: nonprofit rates turned out not to apply, and the working pricing is standard Claude Team, $20/user/month billed annually. This page stays up as the pre-read record. The decision and what happens next live on the decision memo.
The staff diagnostic gives us the starting point, which is further along and less governed than we thought. Full detail on the findings page.
Echidna is choosing which of the tools already in daily use get a governed home, and what happens to the usage and users that don't have one.
Twelve of thirteen working in personal accounts, with no contract protections, no admin visibility, and grantee work sitting in consumer tools. This clarifies the stakes: the risk in this decision sits in not changing status quo.
One tool for everyone, inside the Google Workspace Echidna already pays for and runs on. Outside personal accounts close down, and the policy and tooling decision chooses Gemini as the single approved tool.
This is the lightest footprint to administer: one set of settings, one admin, one data agreement, one place to audit. For a small team with no internal IT function, that simplicity is an obvious asset, and it was the instinct behind the original single-tool brief.
Already paid for, already inside the Workspace data protections, already on 8 work accounts.
Ten people's workflows and history live on ChatGPT and Claude today, so this pathway earns its simplicity only with a dedicated migration plan including real training to move workflows over. Caveat: Some advanced workflows probably won't survive the move (unless the latest reshuffle at Deep Mind leadership result in some massive releases to Gemini feature set).
This is a team that follows rules. The likelier failure mode is quiet compliance: people stop using AI where Gemini can't match what they had or need to accomplish with AI, and the productivity gains and opps the survey shows us simply shrinks. The unsanctioned usage that doesn't stop risks going back under the radar, which recreates Pathway 0 with a policy on top of it (which we have seen happen).
Gemini for everyone as the default: drafting, email, meeting notes, the everyday work inside Workspace. On top of it, a small number of enterprise-grade seats (roughly 4 to 6, for Programs and Grants) on one specialist tool for the heavy synthesis and analysis work. Everything else, including personal accounts for work, closes down.
It matches what the team asked for: 8 of 13 said the right tool depends on the task, and nobody asked for a universal everything-tool.
It gives the heaviest users a governed home instead of a shadow one. The people doing the most ambitious AI work today are doing it on personal paid accounts; this pathway brings that work inside the approved / sanctioned perimeter first.
At nonprofit rates the cost is small: Gemini is bundled into the Workspace editions, and both specialist candidate tools land near $8 per user per month once nonprofit discounts apply (to be confirmed when we apply).
Two tools means more admin overhead: settings to watch on both, features to review as they launch, two vendor relationships. It needs an owner inside Echidna, but we have a proposal for who on the findings page.
Two tiers can read as two classes of staff if we handle it poorly. The framing that keeps it fair: seats follow the work, and the list gets revisited at each checkpoint or on request.
Both real pathways will cost something to the org. Pathway A spends the overhead on migration and training, and accepts some adoption loss. Pathway B spends it on administration, and accepts a second tool to administer. The big question for the group is which overhead Echidna would rather carry, which is a judgment call you are best placed to make.
| DIMENSION | TODAY (P0) | GEMINI ONLY (PA) | BASE + SPECIALIST (PB) |
|---|---|---|---|
| Grantee-data exposure | High consumer tools, no contracts | Low once migration completes | Low enterprise terms on both |
| Shadow-usage risk | High it is the operating model | Medium displaced workflows may go quiet | Low heavy users get a sanctioned home |
| Adoption and morale | Medium energy high, unsupported | Medium some capability quietly lost | Low risk matches stated preferences |
| Admin burden | None and that is the problem | Low one surface | Medium two surfaces, named owner needed |
| Migration and training lift | None | High ten people's workflows move | Medium everyone moves accounts, few move tools |
| Annual tool cost | $0 to Echidna, addtl cost carried by 6 staff | Bundled in Workspace | Bundled + roughly $400–600/yr at nonprofit rates |
Cost figures assume nonprofit pricing: Gemini bundled into Workspace nonprofit editions, and specialist seats near $8/user/month after nonprofit discounts (both major vendors offer them). We can submit both nonprofit applications this week regardless so we can compare real current rates.
If the group lands on Pathway B, the specialist seat comes down to Claude and ChatGPT: both have enterprise tiers with no-training-models defaults, both land near the same nonprofit price, and both already have real usage on the team. One important bias caveat here: Sara and I both use Claude in our own work. So rather than ask you to trust our taste, we'll put the decision on two things you can check.
The current staff sentiment leans one way. Two respondents flagged OpenAI's privacy posture in their scenario answers, and none flagged Claude's. That is signal about trust inside the team, and trust is what a specialist tool has to earn to be used. It is one input, and the bake-off and the vendor answers should ideally be the deciders.
Whichever pathway we choose, we'd frame it as a measured risk with a review cycle built in. At ninety days we look at three things together: actual usage by tool and task, any incidents (with the blameless reporting the policy will establish), and whether staff confidence has shifted at all. If the evidence says adjust, we adjust. The policy will include this.
Whether Echidna would rather spend the overhead on administration (two *or three, including Grammarly* governed tools) or on migration (one tool, more training, some lost workflows). Everything else, including the specialist choice, we can settle together in the room with the bake-off.
Prepared by SA Advisory (Shabnam + Sara), August 2026, as the pre-read for the August 21 tool decision session. Usage figures from the 13 staff diagnostic responses of August 3–12; the survey remains open. Pricing reflects published nonprofit rates as of this week and gets confirmed once Echidna's nonprofit applications are in. The retention case summary draws on court orders and OpenAI's public response in the New York Times litigation, 2025–26. Please shout if anything looks off.