On August 21 the senior team chose Pathway C with Claude as the second tool: every member of the team gets a governed Claude seat alongside Gemini, with an opt-out for anyone who won't use it. This page records what was decided and why, corrects the pricing from the pre-read, and lays out what happens next.
DECIDED · AUG 21
The room went one better than our recommendation. The pre-read recommended Pathway B: Gemini as the base plus a small set of specialist seats. Dana proposed extending it to the whole team, and the room agreed quickly: half the team would have needed access as guides anyway, a split creates two classes of staff, and several people who have held back from other tools out of caution would use them if given a governed home. That instinct deserves the last word:
"I do not use other tools much because I am cautious about the implications of using them. But if I had access, I would use them."
So the shape is: Gemini stays as the base for everyone, since it's bundled into Workspace, takes minutes to administer, and remains the only tool that can write directly to files in Drive. Claude Team seats go to the full team for the heavier synthesis, analysis, and document work. Anyone who genuinely won't use a Claude seat can opt out, and personal accounts for work use close down as part of the migration, with an offboarding path so nobody loses their saved work.
We ran the bake-off as promised: ChatGPT and Claude, identical prompts, web access off, on three real Echidna tasks, ie. a page-cited summary of the Girls' Education Roadmap report, a site-visit pre-read built from a grantee impact report and Echidna's own site-visit guidance, and a grants-data question on a planner extract, followed by live probes designed to make each tool fabricate or fold. The room scored outputs on paper before knowing which tool produced which.
On accuracy, the result was close to even. Both tools handled the tasks credibly, and neither opened a gap the room considered meaningful. That is a genuinely useful finding: it says the accuracy risk the team ranked as its top concern in the diagnostic is managed by working from your own documents and verifying, whichever of the two tools is in the seat. It also meant accuracy could not be the tiebreaker.
The deciding weight was values. With the tools even on performance, the room chose on its read of the two companies behind them: their leadership's public stance on safety and ethics, and how each has handled hard calls. The team's preference for Anthropic's stance was clear and consistent, and it echoes what staff said unprompted in the diagnostic, where two respondents flagged OpenAI's privacy posture and none flagged Claude's. The vendor evidence on retention and legal posture, laid out on the data posture page, showed parity on most axes and sat comfortably with the same choice.
One note on process, since Sara and I disclosed in the pre-read that we both use Claude in our own work: the bake-off existed so this decision would rest on evidence the room could check. The scoring came out even, and the final call was the team's own, made on criteria the team named.
The pre-read quoted nonprofit rates near $8 per seat. In the room, Emily clarified that Echidna is legally an LLC, funded by charitable dollars but without the 501(c)(3) status the nonprofit programs verify against. So the working numbers are standard Claude Team pricing, and they land close to what Emily had budgeted.
Gemini remains bundled into the existing Workspace subscription at no additional cost. Seat counts and the annual-versus-monthly call get finalized during setup, and usage data at the ninety-day checkpoint tells us whether anyone needs a premium seat.
This choice ships as a measured risk with a review built in. At ninety days we look at three things together: actual usage by tool and task, any incidents under the blameless reporting the policy will establish, and whether staff confidence has moved. That checkpoint is also where we revisit the questions the room deliberately parked: whether Gemini usage reflects preference or habit, whether anyone needs a premium seat, and how the two-tool admin load is actually landing. If the evidence says adjust, we adjust.
Confirm the seat count, ie. who is in for launch and who takes the opt-out, so setup and billing start from a real list. Everything else on this page follows from that.
Prepared by SA Advisory (Shabnam + Sara), August 2026, as the record of the August 21 tool decision session with Kim, Dana, Emily, and Katrina. Pricing reflects published Claude Team rates as of this week and gets confirmed at contracting. Please shout if anything looks off.